[ben-adaption]

Adds Adaption as an OpenAI-compatible provider with 7 open-weight models.

• glm-5.2, glm-5.3-flash, qwen3.5-397b, qwen3.8-27b, deepseek-v4-flash, gemma-4-31b-it, gemma-4-26b-a4b-it
• All use base_model; overrides cover only modalities, context, reasoning controls, cache and cost.
• Prices: https://docs.adaptionlabs.ai/inference-stacks/models
• Checked live against the API: structured output, reasoning_content, image input (where listed), implicit prefix caching and max_completion_tokens.

bun validate passes. bun run v1:check lists only the new provider and its 7 models.

[github-actions]

Action items

• [low] [possible mistake] providers/adaption/models/deepseek-v4-flash.toml:16 (also glm-5.2.toml:16, glm-5.3-flash.toml:16) - Check: Costs must be USD/MTok list prices from Adaption's own price list, with the source in a leading comment block above schema = 2 (AGENTS.md → Cost / Comments in TOML; conversion from non-USD requires a noted rate/date). Why: None of the seven hand-authored files carry a top-of-file source comment, and the cited PR URL cannot be verified here. Several values look derived rather than listed (0.04444 / 0.001414 / 0.13332 — input and output sit in an exact ×1.111 relationship), and the full glm-5.2 set (1.4 / 0.26 / 4.4) and glm-5.3-flash set (0.15 / 0.03 / 0.5) exactly match Together's serverless rates in this repo, so the numbers may have been taken from a peer provider or converted from another currency rather than read from Adaption's own USD price page; either would publish incorrect prices or an unnoted conversion. Action: Confirm each price against https://docs.adaptionlabs.ai/inference-stacks/models as USD per MTok (if the listed prices are non-USD, convert and note rate/date per AGENTS.md), and add a leading comment block above the first key in each model file citing the price source and access date.