[mikkel]

Merge order: #27 first, then this PR (#26). This branch already contains #27 at 810c1ba4, brought in with a normal merge commit (no force push). Overlapping lines use #27's wording. Trial merges of #26 onto main + #27 are clean whether #27 lands as a merge commit or a squash, and the result tree is identical to this branch's head. After #27 lands, this PR's diff reduces to the model-name changes below.

Docs pass after #24 made deepseek/deepseek-v4.1-flash-1017 the default. The README and docs still named models that NanoGPT, the default base URL, does not list. This changes docs only, plus one docstring example.

Every id was checked against NanoGPT's free GET /api/v1/models?detailed=true listing, refetched 2026-10-21 19:05 MT. No completion calls were made.

Old (not listed)	New (listed, $ per 1M input / output tokens)	Where
gemini-3-pro-preview (called "the default")	deepseek/deepseek-v4.1-flash-1017, $0.10 / $0.60, the real DEFAULT_MODEL	quickstart, README table and exports, installation, docs/api.md
gpt-4o / claude-sonnet-4-20250514 (complex)	xiaomi/mimo-v2.6-pro ($0.435 / $0.87) / anthropic/claude-sonnet-4.6 ($3 / $15)	README table, docs/cli.md options
gpt5-mini (smartapply)	openai/gpt-5-mini, $0.25 / $2, the listed id for that model	README, installation, docs/api.md (the GPTDIFF_SMARTAPPLY_* lines in docs/cli.md / docs/api.md now come from #27)
gemini-2.0-flash (simple/cheap)	gemini-2.5-flash-lite, $0.10 / $0.40	quickstart tip, docs/cli.md, troubleshooting, docs/api.md retry example
--model deepseek-reasoner	--model deepseek/deepseek-v4.1-flash-1017	docs/examples/basic.md, advanced.md, automation.md ×3, core.py docstring example

Wording. --model is described as taking "a compatible text-generation model" (README and docs/cli.md). It is not "any listed id", because the list also includes decision models such as jaredpalmer/kev-4b. I removed quality claims I couldn't verify for the new ids ("fast and reliable", "most cost-effective", "works great for most tasks"). The README table now gives each model's role and listed price, with a note on where the ids and prices come from and when they were checked. The set_env example in docs/api.md uses xiaomi/mimo-v2.6-pro so that it still shows overriding the default with a different model.

Unchanged on purpose:

• Unit tests that pass "deepseek-reasoner" as an arbitrary model string.
• The patchling/env.py comment that explains why the old default was replaced.
• CHANGELOG.md.
• The CLI's code fallback for --applymodel, openai/gpt-4.1-mini. It is listed ($0.40 / $1.60) and is code, not docs.
• agents/fix-test-agent.py, which uses model="o3-mini" (not listed). It's an example script, not docs, so it's left for a separate change.

Overlap with #27 (resolved in favor of #27)

• docs/api.md SMARTAPPLY paragraph and docs/cli.md GPTDIFF_SMARTAPPLY_MODEL line: docs(cli): describe the real smartapply model fallback (openai/gpt-4.1-mini) #27's wording, which says the variables are CLI-only, --applymodel beats GPTDIFF_SMARTAPPLY_MODEL, the fallback is openai/gpt-4.1-mini, and there is never a fallback to GPTDIFF_MODEL.
• README env line (GPTDIFF_SMARTAPPLY_MODEL): docs(cli): describe the real smartapply model fallback (openai/gpt-4.1-mini) #27's CLI-only row. It merged without conflict.
• A follow-up commit keeps docs: replace delisted model names with ids on NanoGPT's model list #26's own smartapply hints consistent with that. The docs/api.md Model Selection bullet says to pass openai/gpt-5-mini as model= (the CLI reads GPTDIFF_SMARTAPPLY_MODEL), and the README model table says the same.
• No line in README, docs/api.md or docs/cli.md claims GPTDIFF_SMARTAPPLY_MODEL falls back to GPTDIFF_MODEL.

Checks (local)

• Every model id named in README.md and docs/ is on the 2026-10-21 listing. No gemini-3-pro-preview, gpt5-mini, gemini-2.0-flash, gpt-4o or deepseek-reasoner is left in the built HTML.
• mkdocs build --strict passes.
• pytest, offline with the fake_tiktoken stub: 207 passed, 2 skipped, the same per test as main.

No version bump, tag, or publish.