aicodingmodels.csv

Month,Model Name,Company,URL,Description
October 2026,Gemini 4 Argon,Google DeepMind,https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/?utm_source=ai-coding-landscape,"announced September 30, 2026 for trusted cyber defenders through Google's Fairwind Program, then generally available October 13; Google's new frontier model for real-world coding, enterprise knowledge work, and cyber defense; raises the single-response output limit from 64K to 1M tokens; new state of the art on DeepSWE v1.1 at 77.9%, ahead of Claude Opus 5.5 (74.2%) and GPT-6 Astra (74.1%), though it trails both on Terminal-Bench 4.0 at 57.4%; ties for first on CWE-bench v1 at 68% and ranks #1 on the Vals Index; available through the Gemini API in Google AI Studio, in Google Antigravity for Google AI Ultra subscribers, and in the Gemini app for Ultra and Pro subscribers; introductory pricing of $2/$10 per million input/output tokens, rising to $4/$20 when the introductory period ends"
October 2026,Muse Wonder,Meta Superintelligence Labs,https://research.meta.ai/blog/muse-wonder?utm_source=ai-coding-landscape,"October 12, 2026; Meta's new frontier model family for personal and professional agents, memory, and coding, previously code-named Watermelon and trained on Meta's gigawatt-scale Prometheus data center; trained to try other approaches on long-running tasks instead of stalling; Meta reports the larger Muse Wonder Full at 79% on DeepSWE v1.1, ahead of Gemini 4 Argon at 77.9%, in its own testing; 1M-token context; $4/$20 per million input/output tokens with cached input at $0.20; available through Muse Code and the Meta Model API, with Muse Wonder Full limited to Muse Code subscribers"
October 2026,Mistral Large 4,Mistral AI,https://mistral.ai/news/mistral-large-4/?utm_source=ai-coding-landscape,"October 6, 2026 (API preview); open-weight, natively multimodal MoE with 1.05T total / 52B active parameters and a 1.6B vision encoder, trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters; 49.8% on the Artificial Analysis Coding Agent Index, ahead of DeepSeek V4 Pro (0813) and Qwen3.8-Max; ranks in the top five on the Artificial Analysis Cyber Index; $0.68/$2.09 per million input/output tokens at launch; weights due October 27, and Mistral says the preview benchmarks may change before then"
October 2026,Beam,Reflection AI,https://reflection.ai/blog/introducing-beam?utm_source=ai-coding-landscape,"October 5, 2026; the lab's first open-weight model, a sparse MoE with 501B total / 23B active parameters built for coding, reasoning, and agentic work; pretrained on 23.8 trillion tokens, then trained with reinforcement learning on 10,500 NVIDIA GB300 GPUs for four weeks across more than 100 million rollouts; Reflection says it matches GLM-5.2 on reasoning with 3–4x less inference compute and approaches Qwen3.8-Max on coding and agentic tasks; waitlist access now, with Apache 2.0 weights and a technical report due later in October"
September 2026,GPT-6.1 Sol,OpenAI,https://openai.com/index/introducing-gpt-6-1-sol/?utm_source=ai-coding-landscape,"September 29, 2026, announced at DevDay; an upgrade to GPT-6 Sol that nearly matches GPT-6 Astra on agentic coding, computer use, and professional work at one-fifth of Astra's price; 75.2% on DeepSWE v1.1 at high effort for $0.65 per task, against Astra's best of 74.1% at $4.43; 54.0% on Terminal-Bench 4.0 at xhigh effort; 1.05M-token context and 128K max output; cuts responses with factual errors from 11.4% to 7.7%; the new default model in Codex and rolling out across GitHub Copilot; $2/$10 per million input/output tokens, with cached input at $0.10"
September 2026,Claude Sonnet 5.5,Anthropic,https://www.anthropic.com/claude-sonnet-5-5?utm_source=ai-coding-landscape,"September 28, 2026; second model in the Claude 5.5 family, built for well-scoped everyday coding, bug fixing, and polished documents, slides, and spreadsheets; 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5, and 55.5% on CursorBench 4.0; generates output more than 30% faster than Sonnet 5, making it the fastest Sonnet yet, and costs up to 30% less per task; 1M-token context and 128K max output; $2/$10 per million input/output tokens, the same as Sonnet 5; available on all Claude platforms, including AWS, Google Cloud, and Microsoft Azure"
September 2026,Claude Opus 5.5,Anthropic,https://www.anthropic.com/claude-opus-5-5?utm_source=ai-coding-landscape,"September 22, 2026; first model in the Claude 5.5 family, leading Anthropic's benchmarks in agentic coding, computer use, and knowledge work; beats Claude Fable 5.1 on every row of the launch table at 40% of Fable 5.1's price; 66.4% on Terminal-Bench 4.0, ahead of GPT-6 Astra at 57.9%; 1M-token context and 128K max output; $4/$20 per million input/output tokens, 20% below Opus 5, with cache reads at $0.20 and a faster mode at $8/$40; first Anthropic release tested before launch by outside evaluators, including METR"
September 2026,GPT-6 Sol,OpenAI,https://openai.com/index/introducing-gpt-6-sol-and-luna/?utm_source=ai-coding-landscape,"September 22, 2026; balanced GPT-6 model for interactive and agentic coding, trained with the same methods as GPT-6 Astra; available in the API, Codex, ChatGPT, and GitHub Copilot; $2/$10 per million input/output tokens, half the GPT-5.6 Sol rate"
September 2026,GPT-6 Luna,OpenAI,https://openai.com/index/introducing-gpt-6-sol-and-luna/?utm_source=ai-coding-landscape,"September 22, 2026; lightweight, low-cost GPT-6 model for smaller and faster coding tasks; available to free users in the ChatGPT desktop app and in Codex and GitHub Copilot; $0.10/$0.50 per million input/output tokens, the cheapest model in the GPT-6 family"
September 2026,Grok 4.7,xAI,https://x.ai/news/grok-4-7?utm_source=ai-coding-landscape,"September 21, 2026; xAI's most capable model for coding and knowledge work, built on a new 2.1-trillion-parameter base model, up 40% from Grok 4.6, with a longer reinforcement learning run weighted toward multi-hour tasks; 46.3% on CursorBench 4.0 and 71.0% on DeepSWE v1.1 at high effort; available in Cursor, Grok Build, GitHub Copilot, and the xAI API; $2/$6 per million input/output tokens, unchanged from Grok 4.6"
September 2026,Fugu Max,Sakana AI,https://sakana.ai/fugu-max-release/?utm_source=ai-coding-landscape,"September 11, 2026; released with Fugu Ultra v2; an orchestration engine, not a single model, that coordinates a swappable pool of open and specialized models; tops six benchmarks, including Terminal-Bench 2.1 and SWEFish; $2/$6 per million input/output tokens, 40–60% below the output price of comparable frontier models"
September 2026,DeepSeek V4.1 Flash,DeepSeek,https://www.deepseek.com/en/news/deepseek-v4-1-flash/?utm_source=ai-coding-landscape,"September 10, 2026; open-weight multimodal MoE model with 552B parameters, activating 8B on input and 16B on output; DeepSeek says it beats the much larger V4 Pro on performance, cost, speed, and total runtime; 1M-token context and 384K max output; native image and text input"
September 2026,GPT-6 Astra,OpenAI,https://openai.com/index/gpt-6-astra/?utm_source=ai-coding-landscape,"September 3, 2026; flagship for the hardest end-to-end coding, computer use, and research work; a single dense reasoning model with five effort settings, a 1.05M-token context window, and 128K max output; saturates FrontierMath Tier 4 at 97.6% and ARC-AGI-3 at 99.9%, and hits 100% on ExploitBench; $10/$50 per million input/output tokens, 2.5x the GPT-5.6 Sol rate, with prompts over 272,000 tokens billed at double input; rollout began with enterprise customers in OpenAI's Daybreak cybersecurity program"
September 2026,Gemini 3.8 Flash,Google DeepMind,https://blog.google/technology/google-deepmind/gemini-3-8-flash/?utm_source=ai-coding-landscape,"September 2, 2026; fourth Flash release in under four months, tuned for long-horizon coding and autonomous agents; 89.4% on Terminal-Bench 2.1, 73.7% on DeepSWE v1.1, and 59.0% on OSWorld 2.0, beating Claude Opus 5 on three of Google's published benchmarks; 1M-token context, 64K output, with text, image, audio, video, and PDF input; $0.75/$3.75 per million tokens, unchanged from 3.7 Flash, with both figures set to double on January 1, 2027"
September 2026,Muse Spark 1.3,Meta Superintelligence Labs,https://research.meta.ai/blog/introducing-muse-spark-1-3?utm_source=ai-coding-landscape,"September 2, 2026; frontier proprietary reasoning model with significant coding and agentic gains, built to sustain long-horizon work across multiple workflows in a single thread; 1M-token context with text, image, and video input; ranks #6 of 636 models on the Artificial Analysis Intelligence Index; $1.25/$4.25 per million tokens through Muse Code and the Meta Model API, with a contributor endpoint 10 to 20 times cheaper for developers who let Meta train on their data"
September 2026,Claude Fable 5.1,Anthropic,https://www.anthropic.com/claude-fable-and-mythos-5-1?utm_source=ai-coding-landscape,"September 1, 2026; 55.8% on Terminal-Bench 4.0, 73.4% on CursorBench 3.2.0, and 52.6% on Terminal-Bench-Science 0.1, more than double Fable 5; 1M-token context and 128K max output; cache reads cut 75% to $0.25 per million, lowering typical workload costs about 25% and highly agentic workloads up to 45%; $10/$50 per million input/output tokens; now cleared to identify software vulnerabilities, with roughly 60% fewer cyber safeguard interventions per Claude Code session"
September 2026,Claude Mythos 5.1,Anthropic,https://www.anthropic.com/claude-fable-and-mythos-5-1?utm_source=ai-coding-landscape,"September 1, 2026; the same model as Fable 5.1 with more permissive safeguards for vetted cyberdefenders and life scientists; 60.9% on Terminal-Bench 4.0; available through the Cyber Verification Program and the Life Sciences Verification Program, and now powers Claude Security"
August 2026,Qwen3.8-Flash-Next,Alibaba Cloud,https://qwen.ai/blog?id=qwen3.8-flash-next&utm_source=ai-coding-landscape,"August 26, 2026; open-weight multimodal MoE model with 125B parameters and 6B active per token, and an early preview of the architecture behind the coming Qwen4 family; mixes linear-attention and sparse full-attention layers; 262K-token native context, extendable to 1M; 62.5% on SWE-bench Pro, ahead of Claude Opus 4.6 Max at 53.4% and the best open-weight score on that benchmark; released under the Qwen Community License 1.0"
August 2026,GLM-5.3-Flash,Z.ai (Zhipu AI),https://docs.z.ai/guides/vlm/glm-5.3-flash?utm_source=ai-coding-landscape,"August 26, 2026; the stealth ""OX Alpha"" model that topped coding leaderboards anonymously, revealed and open-sourced; 320B total / 18B active natively multimodal hybrid-attention MoE, the first multimodal model in the GLM-5 family, with a 1,048,576-token context window and image and video input; beats GLM-5.2 across benchmarks at roughly one-tenth the price, landing within half a point of Claude Opus 4.8 on Z.ai's internal coding benchmark; $0.075/$0.25 per million tokens; MIT license, weights on Hugging Face"
August 2026,GLM-5.3,Z.ai (Zhipu AI),https://huggingface.co/zai-org/GLM-5.3?utm_source=ai-coding-landscape,"API-only flagship released August 14, 2026, with weights following August 28; focused on software engineering, autonomous agents, and cybersecurity; same MoE architecture as GLM-5.2 at roughly 750B total parameters, with Z.ai attributing the gains to post-training on real-world expert workflows rather than a larger model; 1M-token context and 128K output; 50% coding improvement over GLM-5.2 on its internal Code Bench, 66.9 on DeepSWE v1.1 (up from 46.2), and 84.5% on CyberGym, narrowly ahead of Claude Mythos 5; API access through the GLM Coding Plan starting at $18/month; breaking with GLM-5.2's MIT license, ships under a bespoke ""glm-5.3"" license requiring any Model-as-a-Service business clearing $10B in trailing 12-month revenue to pass a Z.ai security review"
August 2026,Qwen3.8-27B,Alibaba Cloud,https://qwen.ai/blog?id=qwen3.8-27b&utm_source=ai-coding-landscape,"August 14, 2026; dense 27B open-weights model accepting text, image, and video, the smaller companion to Qwen3.8-Max; native 262,144-token context extensible to 1M; outperforms Claude Opus 4.6 Max on 16 of the 24 benchmarks in its official model card, with the largest wins in agentic coding (61.7 vs 53.4 on SWE-bench Pro), instruction following (79.5 vs 62.5 on IFBench), and computer use (84.3 vs 72.7 on OSWorld-Verified); 90.3% on LiveCodeBench v6; Apache 2.0 license"
August 2026,DeepSeek V4 Pro (0813),DeepSeek,https://api-docs.deepseek.com/updates/?utm_source=ai-coding-landscape,"August 13, 2026; V4 Pro leaves preview for general availability across the app, web, and API; adds three reasoning-effort levels and native OpenAI Responses API compatibility, with endpoint identifiers unchanged; 1M-token context with output up to 384,000 tokens; tiered pricing took effect August 16, with off-peak usage at half the peak rate"
August 2026,Grok 4.6,xAI,https://x.ai/news/grok-4-6?utm_source=ai-coding-landscape,"August 12, 2026; built for long-running agents, agentic coding, and interactive visual work, with a 500K-token context window and a new xhigh effort tier; big jumps over Grok 4.5 on DeepSWE (54 to 65.9) and APEX-Agents (47.1 to 57.5), scoring 61 on the Artificial Analysis Intelligence Index to match GPT-5.6 Sol; LiveBench agentic coding regressed, 56.5 to 54.2, the one measured step backward; $2/$6 per million tokens with cached input at $0.50, rising to $4/$12 for prompts of 200K or more"
August 2026,Muse Spark 1.2,Meta Superintelligence Labs,https://ai.meta.com/blog/muse-spark-1-2-muse-code/?utm_source=ai-coding-landscape,"August 5, 2026; a coding-focused release shipped alongside Muse Code, Meta's terminal coding agent, co-trained with the model it runs on; 82.9% on Terminal-Bench 2.1 and 59.3% on DeepSWE 1.1, second only to Claude Opus 5 on Meta's own comparison table, though 1.2 was benchmarked inside Muse Code while 1.1 was benchmarked on mini-swe-agent, so the gain reflects model plus harness; Meta said on August 10 that open weights are coming, without a date"
August 2026,Qwen3.8-Max,Alibaba Cloud,https://qwen.ai/blog?id=qwen3.8&utm_source=ai-coding-landscape,"August 3, 2026, following the July 19 WAIC preview; 2.4T-parameter multimodal flagship for autonomous long-horizon coding and knowledge work, with vision as a continuous feedback loop for planning, execution, and self-correction rather than static input; Alibaba highlights 10+ day runs from an empty repository to production, 500+ turns of chip-design optimization, and 365 days of simulated e-commerce strategy; $2.00/$6.00 per million tokens with implicit caching at $0.25; open weights followed August 12 as Qwen3.8-2.4T-A95B, the first downloadable Max-class Qwen, 2.4T total with 95B active, though the checkpoint is text-only with thinking forced on and ships under a custom license rather than the Apache 2.0 used for the 27B sibling"
July 2026,Claude Opus 5,Anthropic,https://www.anthropic.com/news/claude-opus-5?utm_source=ai-coding-landscape,"July 24, 2026; new default on Claude Max and strongest model on Claude Pro, approaching Fable 5's frontier intelligence at half the price; new state of the art on Frontier-Bench and GDPval-AA, more than doubling Opus 4.8's Frontier-Bench score; on CursorBench 3.2 at max effort, comes within 0.5% of Fable 5's peak at half the cost; three times the next-best model on ARC-AGI 3; remains behind Mythos 5 on biology research and exploit development, though close on vulnerability identification; Anthropic's automated behavioral audit found it the most aligned model to date; $5/$25 per million tokens, with Fast mode at 2.5x speed for double the price"
July 2026,Kimi K3,Moonshot AI,https://www.kimi.com/blog/kimi-k3?utm_source=ai-coding-landscape,"July 16, 2026; the world's first open 3T-class model, 2.8T total parameters on a new Kimi Delta Attention and Attention Residuals architecture, with native vision and a 1M-token context window; Stable LatentMoE activates 16 of 896 experts for roughly 2.5x better scaling efficiency than Kimi K2; Moonshot says it trails Claude Fable 5 and GPT-5.6 Sol overall but leads on several benchmarks, including 88.3% on Terminal-Bench 2.1, 81.2% on FrontierSWE, and 91.2% on BrowseComp; $0.30/$3.00/$15.00 per million tokens for cache-hit input, cache-miss input, and output; weights released July 26 on Hugging Face at ~594GB in native MXFP4, under a custom ""kimi-k3"" license requiring Model-as-a-Service businesses over $20M in trailing 12-month revenue to sign a separate commercial agreement; Moonshot calls it ""open weight,"" not ""open source"""
July 2026,Inkling,Thinking Machines Lab,https://thinkingmachines.ai/news/introducing-inkling/?utm_source=ai-coding-landscape,"July 15, 2026; the lab's first release, an open-weights MoE transformer with 975B total / 41B active parameters, up to 1M-token context, pretrained on 45 trillion tokens across text, images, audio, and video; supports controllable thinking effort and native multimodal reasoning rather than optimizing for a single benchmark; 77.6% on SWE-Bench Verified, 54.3% on SWE-Bench Pro Public, and 63.8% on Terminal-Bench 2.1, reaching Nemotron 3 Ultra's Terminal-Bench score at roughly one-third the tokens; shipped with a preview of Inkling-Small, a 276B/12B-active sibling that matches or exceeds it on several benchmarks; full weights including an NVFP4 checkpoint on Hugging Face, fine-tunable on the lab's Tinker platform, with API access via Together, Fireworks, Modal, Databricks, and Baseten"
July 2026,"GPT-5.6 (Sol, Terra, Luna)",OpenAI,https://openai.com/index/gpt-5-6/?utm_source=ai-coding-landscape,"July 9, 2026; three-tier family where the number marks the generation and each name marks a durable capability tier that can advance on its own schedule; Sol is the flagship for complex coding, cybersecurity, and long-running agentic work at $5/$30 per million tokens; Terra is the balanced everyday tier, competitive with GPT-5.5 at about half the cost, at $2.50/$15; Luna is the fastest and cheapest at $1/$6; Sol with max reasoning sets a new state of the art on the Artificial Analysis Coding Agent Index at 80, plus new highs on Terminal-Bench 2.1 and DeepSWE; introduces max reasoning effort and an ultra mode that runs subagents concurrently; all three tiers are classified at OpenAI's ""High"" risk level for cyber and biological/chemical capability"
July 2026,Muse Spark 1.1,Meta Superintelligence Labs,https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/?utm_source=ai-coding-landscape,"July 9, 2026; multimodal reasoning model for agentic tool use, computer use, and coding, with a 1M-token context window and support for orchestrating multi-agent systems as either primary agent or subagent; handles large codebases including bug diagnosis, feature implementation, and large-scale migrations; 61.5% on SWE-Bench Pro, 80.0% on Terminal-Bench 2.1, and 53.3% on DeepSWE 1.1; leads its benchmark comparison on several agentic evals, including 88.1% on MCP Atlas, 54.7% on JobBench, and 62.1% on Humanity's Last Exam with tools, ahead of Gemini 3.1 Pro, Claude Opus 4.8, and GPT-5.5; launched with the public preview of the Meta Model API, Meta's first paid developer offering, at $1.25/$4.25 per million tokens; early partners include Replit, Cline, and Box"
July 2026,Grok 4.5,Cursor and SpaceXAI,https://cursor.com/blog/grok-4-5?utm_source=ai-coding-landscape,"July 8, 2026; MoE model trained jointly by the two companies, Cursor's first built for more than software engineering, spanning data science, finance, legal work, and other knowledge domains; trained on trillions of tokens of Cursor usage data plus reinforcement learning on difficult problems built by a distributed agent system; new cybersecurity safeguards detect and block bad actors outright rather than silently downgrading intelligence; available in Cursor across desktop, web, iOS, CLI, and SDK; $2/$6 per million tokens for the base model, $4/$18 for the fast variant; a distinct model class from Composer 2.5, which Cursor continues to offer separately"
July 2026,SWE-1.7,Cognition,https://cognition.com/blog/swe-1-7?utm_source=ai-coding-landscape,"July 8, 2026; proprietary software-engineering model trained on a Kimi K2.7 base with a refined reinforcement-learning recipe, optimized for long-horizon asynchronous coding; 42.3% on Cognition's own FrontierCode benchmark and about 81.5% on Terminal-Bench 2.1, which the company positions as within a few points of Claude Opus 4.8 and GPT-5.5; runs at 1,000 tokens per second via Cerebras at $1.97 per task; available inside Devin on Web, Desktop, and CLI; not open source, and the benchmark and cost claims come from Cognition's own launch materials rather than independent testing"
July 2026,Hy3,Tencent,https://www.tencent.com/en-us/articles/2202386.html?utm_source=ai-coding-landscape,"July 6, 2026, following the April 23 preview; 295B total / 21B active MoE with an additional 3.8B-parameter MTP layer, built on a hybrid fast-and-slow-thinking architecture; 57.9% on SWE-bench Pro, 68.3% on SWE-bench Multilingual, 45.6% on NL2repo, and 71.7% on Terminal Bench 2.1, which Tencent says is competitive with GLM-5.1 and GLM-5.2; positioned as a cost-efficient model for software development, office work, financial modeling, front-end design, and game production, with two to five times better parameter efficiency than flagship models; 256K context; Apache 2.0, available day one on Hugging Face and ModelScope, with progressive rollout to OpenRouter, Cline, and Kilo; API live on Tencent Cloud TokenHub"
July 2026,Claude Fable 5,Anthropic,https://www.anthropic.com/news/claude-fable-5-mythos-5?utm_source=ai-coding-landscape,"originally launched June 9, 2026, then suspended under export controls and redeployed July 1; a Mythos-class model made safe for general use via safety classifiers that fall back to Opus 4.8 on cybersecurity, biology/chemistry, and distillation queries; state of the art on software engineering, vision, knowledge work, and long-horizon agentic tasks; scored highest among frontier models on Cognition's FrontierCode evaluation; $10/$50 per million tokens"
July 2026,Claude Mythos 5,Anthropic,https://www.anthropic.com/news/claude-fable-5-mythos-5?utm_source=ai-coding-landscape,"originally launched June 9, 2026, then suspended under export controls and redeployed July 1; the same underlying model as Fable 5 with all safeguards lifted; not generally available, restricted to vetted Project Glasswing cyberdefense and critical infrastructure partners, with a separate trusted access program for biomedical researchers planned; $10/$50 per million tokens"
June 2026,Claude Sonnet 5,Anthropic,https://www.anthropic.com/news/claude-sonnet-5?utm_source=ai-coding-landscape,"June 30, 2026; most agentic Sonnet yet, with performance close to Opus 4.8 at a lower price; default model for Free and Pro plans, also available in Claude Code and the Claude Platform; cyber safeguards enabled by default but substantially weaker cyber capabilities than Opus 4.8 and Mythos 5; introductory pricing $2/$10 per million tokens through August 31, 2026, then $3/$15"
June 2026,LongCat-2.0,Meituan,https://longcat.chat/blog/longcat-2.0?utm_source=ai-coding-landscape,"June 30, 2026; 1.6T-parameter MoE coding and agentic model with ~48B activated parameters per token, a 1M-token context window via LongCat Sparse Attention, and 35T+ pretraining tokens; trained entirely on a 50,000-card domestic AI ASIC cluster with no Nvidia hardware, a first at this scale for non-Nvidia training; previously ran anonymously on OpenRouter as ""Owl Alpha,"" where it led developer usage charts for two months; 59.5 SWE-Bench Pro, 70.8 Terminal-Bench 2.1, 77.3 SWE-Bench Multilingual; MIT licensed; API priced at $0.75/$2.95 per million tokens with no charge for cache hits"
June 2026,Qwen-AgentWorld,Qwen (Alibaba),https://huggingface.co/Qwen/Qwen-AgentWorld-35B-A3B?utm_source=ai-coding-landscape,"June 23, 2026; not a coding model but a world model for simulating coding and agent environments; predicts the next environment state given an agent's action and history across seven domains including Terminal, SWE, Web, and OS; built on Qwen3.5-35B-A3B-Base, 35B total / 3B active, 262K context, Apache 2.0; a larger 397B-A17B variant tops its own AgentWorldBench; useful for training and evaluating coding agents rather than writing code directly"
June 2026,Sakana Fugu,Sakana AI,https://sakana.ai/fugu/?utm_source=ai-coding-landscape,"June 22, 2026; not a standalone model but a multi-agent orchestration system delivered as one OpenAI-compatible API, shipping in two tiers (Fugu and Fugu Ultra); dynamically assembles and coordinates a pool of existing frontier models per task, drawing on two ICLR 2026 papers (TRINITY and Conductor) on learned model orchestration; reports SWE-Bench Pro 73.7 and TerminalBench 2.1 82.1 for Fugu Ultra, though benchmarks compare the orchestrated ensemble against single frontier models and the underlying model pool is proprietary and undisclosed; positioned explicitly as frontier capability without export-control exposure; Fugu Ultra priced at $5/$30 per million tokens; not available in the EU/EEA"
June 2026,GLM-5.2,Z.ai (Zhipu),https://huggingface.co/zai-org/GLM-5.2?utm_source=ai-coding-landscape,"launched June 13, 2026 with MIT open weights released June 16; flagship long-horizon and coding model in the GLM-5 family; 744B-parameter MoE with ~40B active per token and a solid 1M-token context window; introduces IndexShare, which reuses one indexer across every four sparse attention layers to cut per-token FLOPs by 2.9x at 1M context, plus an improved MTP layer that lifts speculative-decoding acceptance length by up to 20%; multiple thinking-effort levels (High and Max); 62.1 SWE-bench Pro, 81.0 Terminal-Bench 2.1, 63.7 ProgramBench; no regional limits"
June 2026,Kimi K2.7 Code,Moonshot AI,https://huggingface.co/moonshotai/Kimi-K2.7-Code?utm_source=ai-coding-landscape,"June 12, 2026; coding-focused agentic model built on Kimi K2.6; 1T total / 32B active MoE with 256K context, native INT4 quantization, and multimodal input (image and video) via a 400M-parameter MoonViT vision encoder; improves long-horizon coding task completion while cutting thinking-token usage roughly 30% versus K2.6; open weights under Modified MIT License"
June 2026,North Mini Code 1.0,Cohere Labs,https://huggingface.co/CohereLabs/North-Mini-Code-1.0?utm_source=ai-coding-landscape,"June 9, 2026; 30B total / 3B active open-weight MoE model optimized for code generation, agentic software engineering, and terminal tasks; 256K context window with 64K max output; decoder-only sparse MoE using interleaved sliding-window and global attention; post-trained with cascaded SFT followed by reinforcement learning with verifiable rewards (RLVR); released under Apache 2.0"
June 2026,Nemotron 3 Ultra,NVIDIA,https://research.nvidia.com/labs/nemotron/Nemotron-3-Ultra/?utm_source=ai-coding-landscape,"June 4, 2026; 550B total parameters, 55B active; open-weight hybrid Mamba-Transformer MoE built for long-running agentic coding and orchestration workflows; scores 48 on the Artificial Analysis Intelligence Index, making it the highest-scoring open-weight model from a US lab at release; 65–70.4% on SWE-bench Verified; up to 5x higher inference throughput than comparable open models; weights, training recipes, and quantized variants all released publicly"
June 2026,MiniMax M3,MiniMax,https://www.minimaxi.com?utm_source=ai-coding-landscape,"June 1, 2026; first open-weight model to combine frontier-level coding, a 1M-token context window, and native multimodal input (text, image, video) in a single architecture; uses new MiniMax Sparse Attention (MSA) design; 59.0% SWE-Bench Pro; API live at launch, open weights committed to Hugging Face and GitHub within ten days"
May 2026,Claude Opus 4.8,Anthropic,https://www.anthropic.com/news/claude-opus-4-8?utm_source=ai-coding-landscape,"May 28, 2026; improvements across coding, agentic tasks, and long-horizon professional work; 88.6% SWE-bench Verified, 69.2% SWE-bench Pro; introduces dynamic workflows in Claude Code enabling hundreds of parallel subagents in a single session; pricing held flat from Opus 4.7 at $5/$25 per million tokens"
May 2026,Qwen3 Coder Next,Alibaba Cloud,https://qwen.ai?utm_source=ai-coding-landscape,"listed May 18, 2026 on LLM Gateway; code-specialist variant in the Qwen3 family designed for long-context repo-scale reasoning"
May 2026,Qwen 3.7 Max,Alibaba Cloud,https://qwen.ai?utm_source=ai-coding-landscape,"announced May 20, 2026 at the Alibaba Cloud Summit; API-only flagship with 1M-token context, native extended-thinking mode, and benchmark wins on SWE-Pro (60.6) and Terminal-Bench 2.0 (69.7); no open weights at launch"
May 2026,Gemini 3.5 Flash,Google DeepMind,https://blog.google/innovation-and-ai/technology/developers-tools/google-ai-studio-io-2026/?utm_source=ai-coding-landscape,announced at Google I/O 2026 (May 19); co-developed with Antigravity and powers Antigravity 2.0
May 2026,Cursor Composer 2.5,Cursor,https://cursor.com/blog/composer?utm_source=ai-coding-landscape,"May 18, 2026; agentic coding model fine-tuned on Moonshot AI's open-weight Kimi K2.5 checkpoint; 79.8% SWE-Bench Multilingual, 69.3% Terminal-Bench 2.0; priced at $0.50 per million input tokens, roughly one-tenth the cost of Claude Opus 4.7 at launch"
May 2026,Gemini 3.1 Flash Lite,Google DeepMind,https://deepmind.google/models/gemini/flash-lite/?utm_source=ai-coding-landscape,"hit AI gateways May 8, 2026; default-tier successor"
May 2026,Zyphra ZAYA1-8B,Zyphra,https://www.zyphra.com?utm_source=ai-coding-landscape,May 2026; 8B MoE trained on AMD hardware
May 2026,GPT-5.5 Instant,OpenAI,https://openai.com?utm_source=ai-coding-landscape,"became ChatGPT default May 5, 2026; replaced GPT-5.4 Instant as the default-tier model hundreds of millions of users interact with daily"
April 2026,DeepSeek V4 Pro,DeepSeek,https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro?utm_source=ai-coding-landscape,"released April 24; 1.6T total / 49B active parameters, new architecture and largest DeepSeek model to date; hybrid thinking/non-thinking; 1M token context window; #2 open-weight reasoning model on Artificial Analysis Intelligence Index; leads open-weight models on GDPval-AA agentic work tasks; MIT license"
April 2026,DeepSeek V4 Flash,DeepSeek,https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash?utm_source=ai-coding-landscape,"released April 24; 284B total / 13B active parameters; DeepSeek's first two-tier lineup with Flash positioned for fast, low-cost inference; one of the cheapest small frontier models available; MIT license"
April 2026,GPT-5.5,OpenAI,https://openai.com/index/introducing-gpt-5-5/?utm_source=ai-coding-landscape,"released April 23; smartest and most intuitive model yet, excels at agentic coding, computer use, and knowledge work; more token-efficient than GPT-5.4 for most Codex tasks while matching per-token latency; 82.7% on Terminal-Bench 2.0"
April 2026,Kimi K2.6,Moonshot AI,https://forum.moonshot.ai/t/meet-kimi-k2-6-advancing-open-source-coding/369?utm_source=ai-coding-landscape,"released April 20; open-weight 1T-parameter MoE with 32B active parameters; first open-weight model to beat GPT-5.4 (xhigh) on SWE-Bench Pro; 80.2% on SWE-Bench Verified, 58.6% on SWE-Bench Pro; supports 4,000+ tool calls and 12+ hour continuous execution; strong cross-language generalization across Rust, Go, Python, and front-end; 256K context window"
April 2026,Claude Opus 4.7,Anthropic,https://www.anthropic.com/news/claude-opus-4-7?utm_source=ai-coding-landscape,"most capable generally available model; scores 87.6% on SWE-bench Verified and 64.3% on SWE-bench Pro; excels at long-horizon agentic tasks complex multi-step coding and professional knowledge work; adds high-resolution vision (2,576px); new adaptive thinking and configurable effort levels including xhigh mode; 1M token context window; first Opus model with automated cybersecurity safeguards"
April 2026,MiniMax M2.7,MiniMax,https://www.minimaxi.com/en/news?utm_source=ai-coding-landscape,released April 13; self-evolving capabilities via production feedback loops; 56.22% on SWE-Bench Pro; most cost-efficient option in its performance tier at ~$0.30/M input tokens; 200K context window
April 2026,Meta Muse Spark,Meta,https://ai.meta.com/muse-spark/?utm_source=ai-coding-landscape,released April 8 from Meta Superintelligence Labs; Meta's first proprietary (non-open-source) model; scores 52 on the Artificial Analysis Intelligence Index (vs. Llama 4 Maverick's 18); leads all models on CharXiv Reasoning at 86.4%; available free on meta.ai in Instant and Thinking modes
April 2026,GLM-5.1,Z.ai (Zhipu AI),https://www.zhipuai.cn/en?utm_source=ai-coding-landscape,"released April 7; 754B MoE; 58.4% on SWE-Bench Pro; Code Arena Elo of 1,530, third globally on agentic web development leaderboard; strong front-end component generation; MIT license"
April 2026,Gemma 4,Google DeepMind,https://deepmind.google/models/gemma/?utm_source=ai-coding-landscape,"four open-weight variants (E2B through 31B Dense) released April 2 under Apache 2.0; 31B Dense scores 80% on LiveCodeBench v6 and 89.2% on AIME 2026; natively multimodal with function calling and agentic workflow support; Codeforces ELO jumped from 110 in Gemma 3 to 2,150"
March 2026,GPT-5.4,OpenAI,https://openai.com/index/introducing-gpt-5-4/?utm_source=ai-coding-landscape,most capable and efficient frontier model for professional work; first mainline model to incorporate GPT-5.3-Codex's coding capabilities; adds native computer use (75% on OSWorld surpassing the 72.4% human expert baseline); 1M token context window; scores 83% on GDPval and ~80% on SWE-bench Verified; rolled out March 5
March 2026,GPT-5.4 mini & nano,OpenAI,https://openai.com/index/introducing-gpt-5-4-mini-and-nano/?utm_source=ai-coding-landscape,smaller efficient variants released March 17; mini approaches GPT-5.4-level coding performance at ~6x lower cost; nano targets classification data extraction and lightweight coding subagents; both optimized for fast iteration in coding workflows
March 2026,NVIDIA Nemotron 3 Super,NVIDIA,https://blogs.nvidia.com/blog/nemotron-3-super-agentic-ai/?utm_source=ai-coding-landscape,120B parameter open-weight hybrid Mamba-Transformer MoE (12B active per token) released at GTC March 11; sets new open-weight record of 60.47% on SWE-bench Verified; 1M token context window; 5x higher throughput than prior generation; fully open — weights datasets and training recipes released
March 2026,Mistral Small 4,Mistral AI,https://mistral.ai/news/mistral-small-4?utm_source=ai-coding-landscape,119B parameter MoE (6B active per token) released March 16; unifies Magistral (reasoning) Pixtral (multimodal) Devstral (agentic coding) and Mistral Small (instruct) into a single model; configurable reasoning effort per request; 256K context window; Apache 2.0 license
March 2026,GLM-5.1,Z.ai (Zhipu AI),https://huggingface.co/zai-org/GLM-5.1?utm_source=ai-coding-landscape,coding-optimized iteration of GLM-5 (744B MoE 40B active) released March 27; achieves 94.6% of Claude Opus 4.6's coding benchmark performance; MIT license; GLM Coding Plan starts at $3/month
February 2026,Gemini 3.1 Pro,Google DeepMind,https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/?utm_source=ai-coding-landscape,upgraded core intelligence for the Gemini 3 series; scores 77.1% on ARC-AGI-2 (more than double Gemini 3 Pro) and 80.6% on SWE-bench Verified; 1M token context window with 65K output; ranked #1 on Artificial Analysis Intelligence Index at launch
February 2026,Claude Sonnet 4.6,Anthropic,https://www.anthropic.com/news/claude-sonnet-4-6?utm_source=ai-coding-landscape,"leads GDPval-AA Elo benchmark for real expert-level work with 1,633 points; preferred over previous Sonnet 70% of the time in Claude Code testing; 1M token context window (beta); default model on Claude.ai free and pro plans and powering GitHub Copilot's coding agent"
February 2026,GPT-5.3-Codex-Spark,OpenAI,https://openai.com/index/introducing-gpt-5-3-codex-spark/?utm_source=ai-coding-landscape,"research preview of OpenAI's first model designed for real-time ultra-fast coding; powered by Cerebras Wafer Scale Engine 3 delivering more than 1,000 tokens per second; 58.4% accuracy on Terminal-Bench 2.0; 128k context window"
February 2026,Qwen3.5,Alibaba Cloud,https://qwen.ai/blog?id=qwen3.5&utm_source=ai-coding-landscape,397B parameter native vision-language model with only 17B active per forward pass via hybrid linear attention and sparse MoE; 76.4% on SWE-bench Verified 52.5% on Terminal-Bench 2; 1M context window; multilingual support expanded from 119 to 201 languages; 8.6x decoding throughput vs. prior generation
February 2026,Zhipu AI GLM-5,Z.ai (Zhipu AI),https://docs.z.ai/guides/llm/glm-5?utm_source=ai-coding-landscape,flagship MoE model with 745B total parameters (44B active) designed for Agentic Engineering; achieves SOTA performance for open-source models narrowing the gap with Claude Opus 4.5; 200k token context window; MIT license; trained on Huawei Ascend infrastructure
February 2026,MiniMax 2.5,MiniMax,https://platform.minimax.io/docs/guides/models-intro?utm_source=ai-coding-landscape,peak-performance model optimized for end-to-end developer workflows including multi-file edits and test-validated repairs; 80.2% score on SWE-Bench; 37% faster than comparable frontier models; 200k context window; thinking mode for complex logic
February 2026,Claude Opus 4.6,Anthropic,https://www.anthropic.com/news/claude-opus-4-6?utm_source=ai-coding-landscape,improved coding skills including better planning sustained agentic tasks operation in larger codebases and enhanced code review and debugging; first Opus-class model with 1M token context window (beta); SOTA on Terminal-Bench 2.0 Humanity's Last Exam GDPval-AA and BrowseComp
February 2026,GPT-5.3-Codex,OpenAI,https://openai.com/index/introducing-gpt-5-3-codex/?utm_source=ai-coding-landscape,most capable agentic coding model combining GPT-5.2-Codex coding performance with GPT-5.2 reasoning capabilities in a single model that's 25% faster; handles long-running tasks involving research tool use and complex execution; first OpenAI model to help create itself
January 2026,SERA-32B,Ai2,https://huggingface.co/allenai/SERA-32B?utm_source=ai-coding-landscape,"first model in Ai2's Open Coding Agents series; achieves 49.5% on SWE-bench Verified; trained using Soft Verified Generation (SVG) 26x cheaper than RL and 57x cheaper than previous synthetic data methods; total training cost approximately $2,000 (40 GPU-days)"
January 2026,Kimi K2.5,Moonshot AI,https://huggingface.co/moonshotai/Kimi-K2.5?utm_source=ai-coding-landscape,"Open-Source Visual Agentic Intelligence; Global SOTA on Agentic Benchmarks: HLE full set (50.2%) BrowseComp (74.9%); Open-source SOTA: MMMU Pro (78.5%) VideoMMMU (86.6%) SWE-bench Verified (76.8%); Agent Swarm (Beta): up to 100 sub-agents 1,500 tool calls 4.5× faster vs. single-agent"
January 2026,GLM-4.7-Flash,Z.ai,https://huggingface.co/zai-org/GLM-4.7-Flash?utm_source=ai-coding-landscape,a local coding and agentic assistant setting a new standard for the 30B class; balancing high performance with efficiency; also recommended for creative writing translation long-context tasks and roleplay
December 2025,M2.1,MiniMax,https://huggingface.co/MiniMaxAI/MiniMax-M2.1?utm_source=ai-coding-landscape,open-source AI model with 10 billion activated parameters (230 billion total); scores 74.0 on SWE-bench Verified and 91.5 on VIBE-Web benchmarks; excels in multi-language programming (Rust Java Go C++ TypeScript) and UI development
December 2025,GLM-4.7,Z.ai,https://huggingface.co/zai-org/GLM-4.7?utm_source=ai-coding-landscape,optimized for AI coding assistance; major improvements over GLM-4.6 including 5.8% gain on SWE-bench and 12.9% on multilingual coding; improvements in UI/webpage generation tool usage and complex reasoning
December 2025,GPT-5.2-Codex,OpenAI,https://openai.com/index/introducing-gpt-5-2-codex/?utm_source=ai-coding-landscape,most advanced agentic coding model for complex real-world software engineering; optimized version of GPT‑5.2 for agentic coding in Codex; improvements on long-horizon work through context compaction stronger performance on refactors and migrations; significantly stronger cybersecurity capabilities
December 2025,Gemini 3 Flash,Google,https://blog.google/products/gemini/gemini-3-flash/?utm_source=ai-coding-landscape,delivers high-speed pro-grade reasoning and outperforms even the Pro model in coding benchmarks; ideal for low-latency agentic workflows and complex multimodal tasks like video analysis and real-time data extraction
December 2025,GPT‑5.2 Thinking,OpenAI,https://openai.com/index/introducing-gpt-5-2/?utm_source=ai-coding-landscape,sets a new state of the art of 55.6% on SWE-Bench Pro; can more reliably debug production code implement feature requests refactor large codebases and ship fixes end-to-end with less manual intervention
December 2025,Devstral 2,Mistral AI,https://mistral.ai/news/devstral-2-vibe-cli?utm_source=ai-coding-landscape,next-generation coding model family in two sizes: Devstral 2 (123B) and Devstral Small 2 (24B); sets the open SOTA for code agents; Devstral 2 ships under modified MIT license Devstral Small 2 under Apache 2.0
December 2025,rnj-1-instruct,Essential AI,https://huggingface.co/EssentialAI/rnj-1-instruct?utm_source=ai-coding-landscape,trained from scratch and optimized for code and STEM with capabilities on par with SOTA open-weight models; strong agentic capabilities (e.g. inside agentic frameworks like mini-SWE-agent); excels at tool-calling
November 2025,Claude Opus 4.5,Anthropic,https://www.anthropic.com/news/claude-opus-4-5?utm_source=ai-coding-landscape,intelligent efficient and the best model in the world for coding agents and computer use; meaningfully better at everyday tasks like deep research and working with slides and spreadsheets
November 2025,GPT-5.1-Codex-Max,OpenAI,https://openai.com/index/gpt-5-1-codex-max/?utm_source=ai-coding-landscape,update to foundational reasoning model trained on agentic tasks across software engineering math research and more; faster more intelligent and more token-efficient
November 2025,Gemini 3,Google,https://blog.google/technology/developers/gemini-3-developers/?utm_source=ai-coding-landscape,most intelligent model delivers unparalleled results across every major AI benchmark; surpasses 2.5 Pro at coding mastering both agentic workflows and complex zero-shot tasks
November 2025,Doubao-Seed-Code,ByteDance Volcengine,https://news.aibase.com/news/22712?utm_source=ai-coding-landscape,achieves breakthroughs in performance price and migration cost; deeply integrated with the TRAE development environment
November 2025,GPT-5-Codex-Mini,OpenAI,https://x.com/OpenAIDevs/status/1986861736041853368?s=20&utm_source=ai-coding-landscape,allows roughly 4x more usage than GPT-5-Codex at a slight capability tradeoff due to the more compact model
November 2025,Mercury Coder,Inception Labs,https://docs.inceptionlabs.ai/get-started/models?utm_source=ai-coding-landscape,dLLM optimized to accelerate coding workflows; streaming tool use and structured output with 128K context window
October 2025,Composer,Cursor,https://cursor.com/blog/2-0?utm_source=ai-coding-landscape,4x faster than similarly intelligent models and built for low-latency agentic coding
October 2025,SWE-1.5,Windsurf Cognition,https://cognition.ai/blog/swe-1-5?utm_source=ai-coding-landscape,a fast-agent frontier-size model with hundreds of billions of parameters that achieves near-SOTA coding performance; 6x faster than Haiku 4.5 and 13x faster than Sonnet 4.5
October 2025,CoDA-1.7B,Salesforce AI Research,https://huggingface.co/Salesforce/CoDA-v0-Base?utm_source=ai-coding-landscape,diffusion-based language model designed for powerful code generation and bidirectional context understanding
October 2025,KAT-Dev-72B-Exp,Kawaipilot,https://huggingface.co/Kwaipilot/KAT-Dev-72B-Exp?utm_source=ai-coding-landscape,an open-source 72B-parameter model for software engineering tasks; achieves 74.6% accuracy on SWE-Bench Verified when evaluated strictly with the SWE-agent scaffold
September 2025,Code World Model (CWM),AI at Meta,https://huggingface.co/facebook/cwm?utm_source=ai-coding-landscape,LLM for code generation and reasoning trained to better represent and reason how code and commands affect the state of a program or system
September 2025,DeepSeek-V3.2-Exp,DeepSeek,https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp?utm_source=ai-coding-landscape,experimental sparse-attention upgrade that halves inference cost while retaining strong code-generation and long-context reasoning
September 2025,GLM-4.6,Z.ai,https://huggingface.co/zai-org/GLM-4.6?utm_source=ai-coding-landscape,features a longer context window superior coding performance advanced reasoning more capable agents and refined writing versus GLM-4.5
September 2025,Claude Sonnet 4.5,Anthropic,https://www.anthropic.com/news/claude-sonnet-4-5?utm_source=ai-coding-landscape,the strongest model for building complex agents the best model at using computers; shows substantial gains on tests of reasoning and math
September 2025,Qwen3-Max-Instruct,Alibaba Cloud,https://qwen.ai/blog?id=241398b9cd6353de490b0f82806c7848c5d2777d&from=research.latest-advancements-list&utm_source=ai-coding-landscape,the official release further elevates its capabilities — particularly in coding and agent performance
September 2025,GPT‑5-Codex,OpenAI,https://openai.com/index/introducing-upgrades-to-codex/?utm_source=ai-coding-landscape,a version of GPT‑5 further optimized for agentic coding in Codex and trained with a focus on real-world software engineering work
September 2025,Kimi K2-Instruct-0905,Moonshot AI,https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905?utm_source=ai-coding-landscape,updated SOTA model with improved agentic and frontend capabilities and increased context length
August 2025,GPT-5,OpenAI,https://platform.openai.com/docs/models?utm_source=ai-coding-landscape,flagship model
August 2025,GPT-5-mini,OpenAI,https://platform.openai.com/docs/models?utm_source=ai-coding-landscape,fast/cost efficient
August 2025,GPT-5-nano,OpenAI,https://platform.openai.com/docs/models?utm_source=ai-coding-landscape,faster/cost efficient
August 2025,Claude Opus 4.1,Anthropic,https://www.anthropic.com/claude/opus?utm_source=ai-coding-landscape,a drop-in replacement for Opus 4
August 2025,Mistral Medium 3.1,Mistral AI,https://x.com/MistralAI/status/1955316715417382979?utm_source=ai-coding-landscape,aka Mistral-Medium-2508 - enterprise-grade model excels in coding tasks
August 2025,Grok Code Fast 1,xAI,https://x.ai/news/grok-code-fast-1?utm_source=ai-coding-landscape,a speedy and economical reasoning model that excels at agentic coding efficient code generation and execution
July 2025,Qwen3-Coder,Alibaba Cloud,https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct?utm_source=ai-coding-landscape,agentic code model
July 2025,Qwen3-Coder-Flash,Alibaba Cloud,https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct?utm_source=ai-coding-landscape,streamlined non thinking agentic code model
July 2025,Kimi K2,Moonshot AI,https://moonshotai.github.io/Kimi-K2/?utm_source=ai-coding-landscape,1 T-param MoE
July 2025,GLM-4.5,Z.ai,https://z.ai/blog/glm-4.5?utm_source=ai-coding-landscape,An open-source LLM designed for intelligent agents
July 2025,Codestral 25.08,Mistral AI,https://mistral.ai/news/codestral?utm_source=ai-coding-landscape,code model for high-precision fill-in-the-middle (FIM) completion
July 2025,Devstral Medium 2507,Mistral × All Hands AI,https://mistral.ai/news/devstral-2507?utm_source=ai-coding-landscape,high-quality and cost-effective model
July 2025,Devstral Small 1.1 2507,Mistral × All Hands AI,https://mistral.ai/news/devstral-2507?utm_source=ai-coding-landscape,agentic model
July 2025,Grok 4,xAI,https://x.ai/grok?utm_source=ai-coding-landscape,trained with reinforcement learning for native tool use including code interpreters making it highly capable for coding and advanced reasoning tasks
June 2025,Gemini 2.5 Pro,Google DeepMind,https://deepmind.google/models/gemini/pro/?utm_source=ai-coding-landscape,flagship model
June 2025,Gemini 2.5 Flash,Google DeepMind,https://deepmind.google/models/gemini/pro/?utm_source=ai-coding-landscape,fast/cost efficient with thinking capabilities
May 2025,Claude Opus 4,Anthropic,https://www.anthropic.com/claude/opus?utm_source=ai-coding-landscape,pushes the frontier in coding agentic search and creative writing
May 2025,Claude Sonnet 4,Anthropic,https://www.anthropic.com/claude/sonnet?utm_source=ai-coding-landscape,improves on Claude Sonnet 3.7 across a variety of areas especially coding
May 2025,DeepSeek-R1-0528,DeepSeek,https://huggingface.co/deepseek-ai/DeepSeek-R1-0528?utm_source=ai-coding-landscape,OSS reasoning model
April 2025,o3,OpenAI,https://platform.openai.com/docs/models?utm_source=ai-coding-landscape,preview reasoning model
April 2025,o4-mini,OpenAI,https://platform.openai.com/docs/models?utm_source=ai-coding-landscape,compact model
April 2025,GPT-4.1,OpenAI,https://platform.openai.com/docs/models?utm_source=ai-coding-landscape,flagship model with 1M token context window
April 2025,Llama 4 Maverick,Meta,https://llama.meta.com/get-started/?utm_source=ai-coding-landscape,code-tuned model
April 2025,Llama 4 Scout,Meta,https://llama.meta.com/get-started/?utm_source=ai-coding-landscape,open-weight model
April 2025,Mellum,JetBrains,https://huggingface.co/JetBrains/Mellum-4b-base?utm_source=ai-coding-landscape,4-B param OSS model
March 2025,DeepSeek-V3-0324,DeepSeek,https://huggingface.co/deepseek-ai/DeepSeek-V3-0324?utm_source=ai-coding-landscape,improved V3 version
February 2025,Gemini 2.0 Flash,Google DeepMind,https://blog.google/technology/google-deepmind/gemini-model-updates-february-2025/?utm_source=ai-coding-landscape,multimodal for high-volume high-frequency tasks
February 2025,Claude 3.7 Sonnet,Anthropic,https://www.anthropic.com/news/claude-3-7-sonnet?utm_source=ai-coding-landscape,first hybrid reasoning model and state-of-the art for coding
February 2025,Grok 3,xAI,https://x.ai/grok?utm_source=ai-coding-landscape,coding capable model
SOURCE,joylarkin/AI-Coding-Landscape,,https://github.com/joylarkin/AI-Coding-Landscape?utm_source=ai-coding-landscape,"The 2026 AI Coding Landscape - Coding agents, CLIs, IDEs, AI app builders, devtools, and more"