feat(wavespeed): add WaveSpeedAI as an OpenAI-compatible LLM provider

[devin-ai-integration]

TLDR

Problem this solves:

• WaveSpeedAI needed manual custom-endpoint config
• No model discovery or cost tracking for it

How it solves it:

• Registers wavespeed as a JSON-configured OpenAI-compatible provider
• Adds all 92 WaveSpeed LLM models to the cost map

Relevant issues

Partially addresses #34618 (LLM API half; media generation is not covered)

Linear ticket

Pre-Submission checklist

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Not captured yet; I don't have a WaveSpeedAI API key, so no live call was made. Once a WAVESPEED_API_KEY is available the proof is

export WAVESPEED_API_KEY=...
curl http://localhost:4000/v1/chat/completions -H "Authorization: Bearer sk-1234" \
  -H "Content-Type: application/json" \
  -d '{"model": "wavespeed/anthropic/claude-opus-4.8", "messages": [{"role": "user", "content": "hi"}]}'

with a proxy config entry of model: wavespeed/anthropic/claude-opus-4.8, then checking the spend log at http://localhost:4000/ui/?page=logs shows a non-zero cost computed from the new cost map entry

Type

🆕 New Feature

Changes

WaveSpeedAI's LLM service is an OpenAI-compatible gateway at https://llm.wavespeed.ai/v1 that also natively exposes /v1/responses and Anthropic /v1/messages, so it fits the existing JSON provider registry rather than needing a Python config class; providers.json gains a wavespeed entry declaring those three endpoints, WAVESPEED_API_KEY, and WAVESPEED_API_BASE, and the usual registration points (LlmProviders, openai_compatible_providers, openai_compatible_endpoints, api-base autodetection in get_llm_provider, provider_endpoints_support.json) are updated to match

WaveSpeed model ids already carry an upstream provider prefix (anthropic/claude-opus-4.8), so a route reads wavespeed/anthropic/claude-opus-4.8 and only the leading wavespeed/ may be stripped; there's a test pinning that, since this is where the OpenRouter-style double-prefix bug would show up

The 92 cost map entries were generated from WaveSpeed's public catalog at https://llm.wavespeed.ai/v1/models, which publishes context windows, per-token prices, cache prices and per-model supported_parameters; those map to max_input_tokens / max_output_tokens, input_cost_per_token / output_cost_per_token, cache_read_input_token_cost / cache_creation_input_token_cost (plus cache_creation_input_token_cost_above_1hr for the Claude 1h cache tier), and the supports_function_calling / supports_tool_choice / supports_response_schema / supports_reasoning / supports_web_search flags. Models with tiered long-context pricing (Qwen, Gemini, MiniMax) get input_cost_per_token_above_256k_tokens-style keys derived from the catalog's conditional_prices, so a 300k-token Qwen 3.6 Flash request bills at the long-context rate

The catalog publishes no vision, pdf, adaptive-thinking or sampling-param flags, so those four are inherited from the canonical cost map entry for the same underlying model when one exists (wavespeed/anthropic/claude-fable-5 picks up supports_adaptive_thinking: true and supports_sampling_params: false from claude-fable-5, which the repo's cross-variant consistency tests require); models with no canonical counterpart simply omit them rather than being guessed from the name. Media generation (/v1/predictions, the Replicate-style async submit/poll API in the issue) is a separate surface and isn't part of this PR

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

Link to Devin session: https://app.devin.ai/sessions/e43399c20a854c7c90fc003efacc40f7

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

Adds WaveSpeedAI as a JSON-configured OpenAI-compatible provider, including provider resolution, Chat Completions, Responses and Messages endpoint registration, model pricing and capability metadata, and focused provider and cost-calculation tests

Confidence Score: 5/5

The PR appears safe to merge; the new provider follows the existing JSON-configured OpenAI-compatible integration paths without an identified actionable defect

Provider resolution, enum and registry wiring, endpoint configuration, model-prefix preservation, and pricing fields are coordinated, and the specialized pricing keys are supported by the existing generic cost-calculation path

Important Files Changed

Filename	Overview
litellm/llms/openai_like/providers.json	Registers WaveSpeedAI through the existing JSON provider framework with the expected credentials, base URL, and supported API surfaces
litellm/litellm_core_utils/get_llm_provider_logic.py	Adds API-base autodetection for WaveSpeedAI while preserving the established provider-prefix resolution behavior
litellm/constants.py	Adds WaveSpeedAI to the OpenAI-compatible endpoint and provider registries
litellm/types/utils.py	Adds the WaveSpeedAI provider identifier to the central provider enum
model_prices_and_context_window.json	Adds WaveSpeed model pricing, context limits, conditional pricing tiers, and capability metadata using fields consumed by the existing cost calculator
litellm/model_prices_and_context_window_backup.json	Mirrors the WaveSpeed model metadata in the local backup cost map
provider_endpoints_support.json	Advertises the three API surfaces declared by the WaveSpeed JSON provider configuration
tests/test_litellm/llms/openai_like/test_wavespeed_provider.py	Covers provider registration, prefix preservation, environment resolution, endpoint URL generation, model capabilities, standard cost calculation, and long-context pricing

_{Reviews (1): Last reviewed commit: "feat(wavespeed): add WaveSpeedAI as an O..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_wavespeed_provider (c6eec29) with litellm_internal_staging (b9b27c2)}