fix: add Databricks models databricks-qwen35-122b-a10b and databricks-qwen3-next-80b-a3b-instruct

[github-actions]

fix: add Databricks models databricks-qwen35-122b-a10b and databricks-qwen3-next-80b-a3b-instruct

Closes #644

Source issue: #644

Summary

Field	Value
Provider	databricks
Primary model	databricks-qwen35-122b-a10b
Changed models	databricks-qwen35-122b-a10b databricks-qwen3-next-80b-a3b-instruct
Added models	databricks-qwen35-122b-a10b databricks-qwen3-next-80b-a3b-instruct
Updated models	None
Verification sources	1 2 3 4 5 6 7 8

Verified metadata

Model	Display name	Parent	Providers	Format	Flavor	Token limits	Pricing	Lifecycle
databricks-qwen35-122b-a10b	Qwen 3.5 122B-A10B		databricks	openai	chat	input=256000, output=8000	n/a	reasoning=true
databricks-qwen3-next-80b-a3b-instruct	Qwen 3-Next 80B-A3B Instruct		databricks	openai	chat	input=262144, output=16384	n/a	active

Verification notes

Verification

Official sources and fields verified

1. https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models (Databricks Supported Models page)

   • databricks-qwen35-122b-a10b: model ID, 256K context window, 8K max output tokens, Public Preview status, text-only input, "reasoning cannot be disabled"
   • databricks-qwen3-next-80b-a3b-instruct: model ID, Public Preview status, text-only input. Token limits NOT published on this page.

2. https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview (Databricks Foundation Model Overview)

   • Both model IDs confirmed in regional availability tables for pay-per-token APIs

3. https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct (upstream Qwen model card)

   • Context Length: 262,144 natively (extensible to 1,010,000 with YaRN)
   • Recommended output length: 16,384 tokens for most queries
   • Used as fallback for databricks-qwen3-next-80b-a3b-instruct since Databricks does not publish specific token limits

4. https://huggingface.co/Qwen/Qwen3.5-122B-A10B (upstream Qwen model card)

   • Confirms reasoning/thinking model with <think> tokens
   • Native context: 262,144, recommended output: 32,768 — Databricks restricts to 256K/8K

sync_models (LiteLLM) cross-check

Neither databricks-qwen35-122b-a10b nor databricks-qwen3-next-80b-a3b-instruct appear in the LiteLLM model_prices_and_context_window_backup.json catalog. No numeric fields to cross-check — both models are absent from sync_models entirely.

Fields not published or not applicable

• Pricing (input_cost_per_mil_tokens, output_cost_per_mil_tokens): Databricks does not publish stable per-model pricing for Foundation Model API models. Omitted.
• Cache pricing (input_cache_read_cost_per_mil_tokens, input_cache_write_cost_per_mil_tokens): Not applicable for Databricks. Omitted.
• multimodal: Both models are text-only per Databricks docs. Omitted (defaults to falsy).
• parent: Neither model is a dated snapshot or location-scoped variant of a stable alias. Omitted.
• deprecation_date: Both are active (Public Preview). Omitted.
• supported_regions: Not applicable — these are Databricks models, not Vertex models.
• locations: Not required for Databricks provider.

Caveats

• databricks-qwen3-next-80b-a3b-instruct token limits (262,144 input / 16,384 output) are sourced from the upstream Qwen HuggingFace model card, not from Databricks documentation. Databricks may impose different limits at the API level. The output limit of 16,384 is described as "recommended" in the model card, not a hard architectural limit.
• databricks-qwen35-122b-a10b upstream model supports 262,144 context and 32,768 output, but Databricks restricts to 256,000 input / 8,000 output per their supported-models page.

sync_models vs proposed update

sync_models cross-check found differences. Official provider verification was used for the applied values, and sync_models discrepancies are listed below for review.

Model	Field	Proposed update	sync_models	sync_models source models
databricks-qwen35-122b-a10b	catalog entry	present	missing	None
databricks-qwen3-next-80b-a3b-instruct	catalog entry	present	missing	None

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ai-proxy	Ready	Preview, Comment	May 24, 2026 2:25am