[bhaveshpatel640]

Add explicit reasoning_effort on account OpenAI, Claude and TrueFoundry Managed models

Provider	Model	YAML	Provider docs	Verdict
OpenAI	gpt-4o	Removed	gpt-4o has no reasoning tokens and no effort values	Match
OpenAI	gpt-5.4-nano	Default none. none, low, medium, high, xhigh	Model page: same. Default none	Match
OpenAI	gpt-5.5	Default medium. none, low, medium, high, xhigh	Model page: same. Default medium	Match
OpenAI	gpt-5.6-luna	Default medium. none, low, medium, high, xhigh, max	Model page: same	Match
OpenAI	gpt-5.6-sol	Default medium. none, low, medium, high, xhigh, max	Model page: same	Match
OpenAI	gpt-5.6-terra	Default medium. none, low, medium, high, xhigh, max	Model page: same	Match
OpenAI	gpt-6-astra	Default medium. low, medium, high, xhigh, max	Model page: those values. none returns HTTP 400. The page does not name a default. The reasoning guide states medium for the other GPT-6 models	Match on values
OpenAI	gpt-6-luna	Default medium. none, low, medium, high, xhigh, max	Model page: same. Default medium	Match
OpenAI	gpt-6.1-sol	Default medium. low, medium, high, xhigh, max	Model page: same. none and minimal are rejected	Match
Anthropic	claude-fable-5	Default high. low, medium, high, xhigh, max	Effort: same. Default high	Match
Anthropic	claude-haiku-4-5-20251001	Removed	Effort is not documented for Haiku 4.5. Haiku 5.5 is the Haiku line that supports it	Match
Anthropic	claude-opus-4-6	Default high. low, medium, high, max	Same. xhigh is not available on Opus 4.6	Match
Anthropic	claude-opus-4-8	Default high. low, medium, high, xhigh, max	Same. Default high	Match
Anthropic	claude-opus-5	Default high. low, medium, high, xhigh, max	Same. Default high	Match
Anthropic	claude-opus-5-5	Default medium. low, medium, high, xhigh, max	Same. Default medium	Match
Anthropic	claude-sonnet-4-6	Default high. low, medium, high, max	Same. xhigh is not available on Sonnet 4.6	Match
Anthropic	claude-sonnet-5	Default high. low, medium, high, xhigh, max	Same. Default high	Match
Anthropic	claude-sonnet-5-5	Default high. low, medium, high, xhigh, max	Same. Default high	Match
TrueFoundry Managed	claude-haiku-4-5-20251001	Removed	Same as the Anthropic Haiku 4.5 row	Match
TrueFoundry Managed	claude-opus-5-5	Default medium. low, medium, high, xhigh, max	Same as the Anthropic Opus 5.5 row	Match
TrueFoundry Managed	claude-sonnet-5-5	Default high. low, medium, high, xhigh, max	Same as the Anthropic Sonnet 5.5 row	Match
TrueFoundry Managed	deepseek-v4-flash-0731	Default high. none, low, high, max	DeepSeek: same. Default high. minimal maps to low. medium and xhigh map to high	Match
TrueFoundry Managed	deepseek-v4-pro-0813	Default high. none, low, high, max	Same DeepSeek values	Match
TrueFoundry Managed	deepseek-v4.1-flash	Default high. none, low, high, max	QwenCloud: default high, low, high, max for this snapshot	Match
TrueFoundry Managed	glm-5.3	Default max. low, high, max	Z.ai: same. Other values error on the API	Match
TrueFoundry Managed	glm-5.3-flash	Default max. low, high, max	Same as GLM-5.3	Match
TrueFoundry Managed	gpt-6-astra	Default medium. low, medium, high, xhigh, max	Same as the OpenAI Astra row	Match on values
TrueFoundry Managed	gpt-6-luna	Default medium. none, low, medium, high, xhigh, max	Same as the OpenAI Luna row	Match
TrueFoundry Managed	gpt-6.1-sol	Default medium. low, medium, high, xhigh, max	Same as the OpenAI 6.1 Sol row	Match
TrueFoundry Managed	gpt-oss-120b	Default medium. low, medium, high	OpenAI: same	Match
TrueFoundry Managed	grok-4.7	Default high. low, medium, high, xhigh	xAI: same. Reasoning cannot be turned off	Match
TrueFoundry Managed	kimi-k2.7-code	Removed	Moonshot: reasoning_effort is not supported. Thinking is always on	Match
TrueFoundry Managed	kimi-k3	Default max. low, high, max	Moonshot: same. Thinking cannot be turned off	Match
TrueFoundry Managed	mimo-v2.6-flash	Default medium. none, minimal, low, medium, high, xhigh, max	Xiaomi: those values, plus ultra, which the catalog enum does not allow. Only none turns thinking off. Every other value enables thinking at the same depth. Xiaomi does not name medium as the default	Accepted values match, except ultra. Default is ours
TrueFoundry Managed	mimo-v2.6-pro	Same as Flash	Same Xiaomi behavior	Same as Flash
TrueFoundry Managed	minimax-m3	Removed	MiniMax: M3 ignores reasoning_effort. Depth tuning is only on M3.1-Flash (low, medium, high, xhigh, max, default max)	Match
TrueFoundry Managed	nemotron-3-ultra	Default high. none, medium, high	NVIDIA: same	Match
TrueFoundry Managed	qwen3.8-max	Default xhigh. low, medium, xhigh	DashScope: same. The Responses API also accepts none and maps high and max to xhigh	Match on the DashScope chat levels
TrueFoundry Managed	text-embedding-3-large	Removed	Embeddings have no reasoning parameter	Match
TrueFoundry Managed	text-embedding-3-small	Removed	Same	Match

---

Note

Low Risk
Catalog-only YAML updates to inference parameters; no application runtime, auth, or billing logic changes.

Overview
This PR aligns model catalog YAML with provider-specific reasoning_effort behavior so callers get explicit defaults, allowed enums, and models that ignore the knob no longer expose it.

Anthropic (claude-fable-5, claude-opus-4-8, claude-opus-5, claude-sonnet-5): adds reasoning_effort default high with low / medium / high / xhigh / max.

TrueFoundry Managed: adds or tightens reasoning_effort on DeepSeek v4 variants (none, low, high, max, default high), Kimi K3 (low, high, max, default max), MiMo v2.6 flash/pro (full ladder including none / minimal, default medium), Nemotron 3 Ultra, and Qwen3.8-Max; supportedValues on gpt-oss-120b (low, medium, high). removeParams: reasoning_effort on Kimi K2.7 Code and MiniMax M3 where the upstream API does not honor it.

These definitions feed catalogue-driven validation and SDK live-test reasoning scenarios.

^{Reviewed by Cursor Bugbot for commit 19fefd4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

SDK live feature checks

Result: failed (11/107 scenarios)

Informational — does not block merge.

Model	Pass	Fail
anthropic/claude-fable-5	23	0
anthropic/claude-opus-4-8	23	0
anthropic/claude-opus-5	25	0
anthropic/claude-sonnet-5	25	0
truefoundry-managed/deepseek-v4-flash-0731	0	1
truefoundry-managed/deepseek-v4-pro-0813	0	1
truefoundry-managed/deepseek-v4.1-flash	0	1
truefoundry-managed/gpt-oss-120b	0	1
truefoundry-managed/kimi-k2.7-code	0	1
truefoundry-managed/kimi-k3	0	1
truefoundry-managed/mimo-v2.6-flash	0	1
truefoundry-managed/mimo-v2.6-pro	0	1
truefoundry-managed/minimax-m3	0	1
truefoundry-managed/nemotron-3-ultra	0	1
truefoundry-managed/qwen3.8-max	0	1

Failures

• truefoundry-managed/deepseek-v4-flash-0731 setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/deepseek-v4-pro-0813 setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/deepseek-v4.1-flash setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/gpt-oss-120b setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/kimi-k2.7-code setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/kimi-k3 setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/mimo-v2.6-flash setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/mimo-v2.6-pro setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/minimax-m3 setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/nemotron-3-ultra setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)
• truefoundry-managed/qwen3.8-max setup: provider not in SDK live harness registry (add providers.ts + GitHub secret)