feat: add Venice AI models and token prices to providers - #26970
feat: add Venice AI models and token prices to providers#26970Kutsukey wants to merge 12 commits into
Conversation
|
|
There was a problem hiding this comment.
Pull request overview
Adds Venice AI model metadata (context windows, capabilities, and per-token pricing) into LiteLLM’s central pricing/context registry so Venice models can be referenced consistently across the codebase.
Changes:
- Added a large set of
venice/*model entries (token limits, pricing, capability flags) to the main pricing/context registry. - Mirrored the same Venice model entries into the backup pricing/context registry.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.
| File | Description |
|---|---|
| model_prices_and_context_window.json | Adds Venice model entries with pricing + context/capability metadata to the primary registry. |
| litellm/model_prices_and_context_window_backup.json | Mirrors the same Venice additions in the backup registry for parity. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| "venice/grok-41-fast": { | ||
| "max_tokens": 30000, | ||
| "max_input_tokens": 1000000, | ||
| "max_output_tokens": 30000, | ||
| "input_cost_per_token": 2.3000000000000002e-07, |
| "max_output_tokens": 32000, | ||
| "input_cost_per_token": 1e-06, | ||
| "output_cost_per_token": 3.2000000000000003e-06, | ||
| "litellm_provider": "venice", |
| "venice/grok-41-fast": { | ||
| "max_tokens": 30000, | ||
| "max_input_tokens": 1000000, | ||
| "max_output_tokens": 30000, | ||
| "input_cost_per_token": 2.3000000000000002e-07, |
| "max_output_tokens": 32000, | ||
| "input_cost_per_token": 1e-06, | ||
| "output_cost_per_token": 3.2000000000000003e-06, | ||
| "litellm_provider": "venice", |
Greptile SummaryAdds 73 Venice AI model entries to the pricing/context-window registry with correct per-token pricing, token limits, and capability flags ( Confidence Score: 5/5Safe to merge; the only outstanding finding is a P2 schema-comment regression. All previous P1 findings (zero pricing, missing vision/reasoning flags, missing trailing newline, MiniMax supports_reasoning removal) have been addressed in this revision. The sole remaining issue is a P2 documentation/schema-description overwrite that does not affect runtime behaviour. No files require special attention beyond the P2 schema comment on line 14 of
|
| Filename | Overview |
|---|---|
| model_prices_and_context_window.json | Adds 73 Venice AI model entries with pricing and capability flags; also fixes duplicate Bedrock/MiniMax entries, adds max_tokens to several audio/Gemini models, removes duplicate supports_max_reasoning_effort JSON keys, and accidentally overwrites the max_tokens schema description. |
| litellm/model_prices_and_context_window_backup.json | Mirror of the main pricing file; receives the same Venice AI additions and ancillary fixes. |
Reviews (7): Last reviewed commit: "fix(venice): fix minimax reasoning regre..." | Re-trigger Greptile
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
|
I checked the official and current Venice AI Model Catalog as well as the /models API endpoint. The model venice/grok-code-fast-1 no longer exists on their platform (it was likely deprecated or renamed to grok-41-fast since the original issue was opened in March). I have added all currently active models to prevent any routing errors. I also fixed the floating-point artifacts and added the missing newlines requested by Greptile. |
- BerriAI/litellm#26972 merge-after-nits: virtual keys filter routed through React Query cache - BerriAI/litellm#26970 needs-discussion: Venice AI additions contaminated by databricks row churn + capability-flag drop - BerriAI/litellm#26968 merge-after-nits: tighten router-settings-override fallback validation + mock-testing strip
| "output_cost_per_token": 2e-07, | ||
| "litellm_provider": "venice", | ||
| "mode": "chat" | ||
| }, | ||
| "venice/arcee-trinity-large-thinking": { | ||
| "max_tokens": 65536, | ||
| "max_input_tokens": 256000, | ||
| "max_output_tokens": 65536, | ||
| "input_cost_per_token": 3.125e-07, |
There was a problem hiding this comment.
venice/claude-opus-4-6-fast is priced at 6× more than venice/claude-opus-4-6 — unusual since "fast" variants are normally cheaper, not pricier. Similarly, venice/openai-gpt-54-pro and venice/openai-gpt-55-pro are priced roughly 12× higher than venice/openai-gpt-4o-2024-11-20. This pattern suggests a possible unit-conversion error (e.g. applying the per-1M divisor twice, or the source already being in per-token units). Please verify these values against the raw Venice API response.
There was a problem hiding this comment.
I verified these values directly against the live Venice API /models response and their official Model Catalog. There is no conversion error. Venice explicitly prices claude-opus-4-6-fast at $36.00/$180.00 per 1M tokens (exactly 6x the base tier) because it is a dedicated/premium routing endpoint. The GPT-5.x Pro pricing also exactly matches their raw API response.
392031a to
ffc55f8
Compare
03ec390 to
9c455bf
Compare
68cdf59 to
83c872a
Compare
|
This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. |
Relevant issues
Fixes #24229
Pre-Submission checklist
tests/test_litellm/directory (Note: This is a JSON data addition, no new logic tests required)make test-unit@greptileaiScreenshots / Proof of Fix
N/A - This is a pure JSON configuration update. Data was fetched directly from the Venice AI API (
/models?type=text) to ensure accurate token limits and correct/1Mpricing conversions.Type
🆕 New Feature
Changes
model_prices_and_context_window.json.input_cost_per_tokenformat.