feat: accept max reasoning effort

[ayush-or]

TL;DR

Adds max to PINNED_REASONING_EFFORTS, so benchmark configs can run models such as openai/gpt-6-luna and deepseek/deepseek-v4.1-flash at their max effort.

What changed?

• src/harness/constants.ts: max is now the first entry in PINNED_REASONING_EFFORTS. That also adds it to REASONING_EFFORTS, the reasoningEffort / userReasoningEffort config schemas, and the CLI --reasoning-effort check.
• reasoningRequestFor("max") returns { effort: "max" }. The SDK's request effort enum already includes max. A test covers it.

Why?

The model catalog lists max for GPT-6 Luna and DeepSeek V4.1 Flash, but configs that use it fail schema validation. DeepSeek V4.1 Flash supports only max/high/low, so sending xhigh runs it at high. The agent CLI efforts (ORI_REASONING_EFFORTS) already include max, and so does toOriReasoningEffort's output type, so the agentic suites now receive it unchanged.

How to test

bun test src/harness/constants.test.ts
bun run src/cli/index.ts --benchmark gpqa_diamond --model openai/gpt-6-luna --reasoning-effort max --limit 1

The second command now passes argument validation and sends reasoning: { effort: "max" }.

Benchmark impact

No change to existing runs or their defaults. max is a new opt-in value.

Reviewer focus

• Whether any consumer of ReasoningEffort should reject max. I found none: toOriReasoningEffort already maps into a superset that includes max.

Checklist

• Tests cover changed behavior
• Public API or configuration changes are backward compatible, or the break is documented
• Benchmark changes document dataset provenance and licensing
• No credentials, private results, or restricted dataset contents are included
• Documentation is updated where needed

Link to Devin session: https://openrouter.devinenterprise.com/sessions/116e7df8fcf449b4b19b1919ffe3cec5
Open in Devin Desktop: https://openrouter.devinenterprise.com/desktop/session/116e7df8fcf449b4b19b1919ffe3cec5?variant=devin
Requested by: @ayush-or

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access that start with 'Devin'.

• Disable automatic comment, CI, and merge conflict monitoring

---

Original prompt from Ayush

SYSTEM:
<latest_message>
Ayush Patel (U0B8L6RNMA9) [ts=1791424352.095979]: @Devin !router_benchmark run these 3 models at these reasoning levels (single model baselines) on these 5 benchmarks. mimo v2.6pro should be at highest thinking level whatever it is
</latest_message>

=== BEGIN THREAD HISTORY (in #agents-benchmarks) ===
Ayush Patel (U0B8L6RNMA9) [ts=1791424343.837149]: !router_benchmark run these 3 models at these reasoning levels (single model baselines) on these 5 benchmarks. mimo v2.6pro should be at highest thinking level whatever it is
no acu,time,spend limit. no smoke, start full 1 epoch each right away.

ATTACHMENT:"https://openrouter.devinenterprise.com/attachments/b4d58163-f1db-4112-b959-ddb4aacb23b6/image.png"

ATTACHMENT:"https://openrouter.devinenterprise.com/attachments/112fbfad-67b2-430b-af7f-e9edbfa52e1a/image.png"

Ayush Patel (U0B8L6RNMA9) [ts=1791424352.095979]: @Devin !router_benchmark run these 3 models at these reasoning levels (single model baselines) on these 5 benchmarks. mimo v2.6pro should be at highest thinking level whatever it is
=== END THREAD HISTORY ===
Channel ID: C0BAKP8P5C3
Thread URL: https://openrouter.slack.com/archives/C0BAKP8P5C3/p1791424343837149?thread_ts=1791424343.837149&amp;cid=C0BAKP8P5C3

The <latest_message> is the message that you should use to guide your goals + task for this session, and you should use the rest of the slack thread as context.
A [ts=...] marker on a Slack message is that message's timestamp. To act on a specific message with the slack tool (e.g. adding an emoji reaction via the reaction command), pass that value as timestamp along with the Channel ID — no extra lookup call is needed.

@playbook:playbook-67bd60e6265449979d13f557104b2ef7