[mikkel]

Bug

DEFAULT_MODEL in 0.3.0 is xiaomi/mimo-v2.5-pro-ultraspeed. NanoGPT has retired that model, and it returns HTTP 410. So any generateDiff/smartapply call made without model and without GPTDIFF_MODEL fails.

Fix

The new default is deepseek/deepseek-v4.1-flash-1017, the same id patchling-py uses (companion PR: 255BITS/patchling-py#24).

Why this model: Mikkel asked for DeepSeek V4.1, the 1017 release. In NanoGPT's free GET https://nano-gpt.com/api/v1/models?detailed=true listing (fetched 2026-10-21 13:03 MT):

• deepseek/deepseek-v4.1-flash-1017 ("DeepSeek V4.1 Flash 1017", the October 17, 2026 snapshot) is listed at $0.10 / $0.60 per 1M input/output tokens, with a 1,048,576 context and 393,216 max output tokens. It is the only id that names the 1017 release with standard serving.
• Not chosen: the same snapshot is also listed as deepseek/deepseek-v4.1-flash-1017:thinking (same price, thinking always on) and as TEE/deepseek-v4.1-flash (secure-enclave serving, $0.65 / $1.45). The undated deepseek/deepseek-v4.1-flash is a rate-limited beta.
• NanoGPT lists no non-Flash V4.1 model. "V4.1" here can only mean V4.1 Flash; the only DeepSeek Pro models are v4.
• Neither retired default is listed: xiaomi/mimo-v2.5-pro-ultraspeed (JS) and deepseek-reasoner (Python).

An earlier commit on this branch used xiaomi/mimo-v2.6-pro. The latest commit replaces it everywhere.

Changes

• src/env.js, landing/lib/env.js (kept byte-identical for landing_lib_sync), regenerated src/env.d.ts (npm run types), the README env table, and the test/live.test.js header comment.
• Verified catalog test/fixtures/demo-models.json:
  • Adds deepseek/deepseek-v4.1-flash-1017, copied field for field from the 2026-10-21 listing.
  • verifiedOn moves to 2026-10-21, and catalogSha256 now covers that listing. The three existing entries (z-ai/glm-5.3-flash, xiaomi/mimo-v2.6-flash, xiaomi/mimo-v2.6-pro) were re-checked and are unchanged.
  • The README notes the re-check.
• New test/default_model.test.js asserts that the default is deepseek/deepseek-v4.1-flash-1017, that the retired id is not the default (in either src or landing/lib), that the default is in the verified catalog and cheap (≤ $0.50 / $1 per 1M), and that the .d.ts literal matches.
• CHANGELOG: note added under [0.3.1] — Unreleased, "Behavior changes".

Unchanged on purpose:

• The filterRelevant default xiaomi/mimo-v2.6-flash is still listed.
• The landing-page demo default and pinned picks are unchanged.
• The retired id stays in landing/index.html's saved-pick migration list and its test, so visitors who saved it are reset.

No version bump, tag, or publish.

Tests (local)

npm test: 248 tests, 244 passed, 0 failed, 4 skipped (live). Main: 240 passed, 4 skipped.

Live smoke (one call, authorized by Mikkel)

One tiny generateDiff call through the default path: no model argument, GPTDIFF_MODEL unset, the default NanoGPT base URL, and max_tokens 400. The task was a one-line edit (hello → hi).

• Model sent and model returned: deepseek/deepseek-v4.1-flash-1017, HTTP 200.
• 128 prompt + 31 completion = 159 tokens, cost $0.0000314, about 2 s.
• The result was a valid unified diff (-hello / +hi).
• The API key came from the environment and was never printed, logged or committed.

[cloudflare-workers-and-pages]

Deploying patchling with    Cloudflare Pages

Latest commit:	5a332b9
Status:	✅  Deploy successful!
Preview URL:	https://57b6f2f6.patchling.pages.dev
Branch Preview URL:	https://fix-default-model-retired.patchling.pages.dev

View logs