[timmoser]

What

Adds a [[models]] entry for gpt-6.1-sol in Versions.toml: $2 input, $0.10 cached input, $10 output and $2.50 cache write per million tokens.

Why

Planning a run on GPT-6.1 Sol failed with pricing missing for codex/gpt-6.1-sol; add a verified pricing entry to Versions.toml before planning.

How the rates were checked

• BenchLM's OpenAI price table (updated 2026-10-08) lists $2.00 / $0.10 / $10.00. Its rows for GPT-6 Astra ($10 / $1 / $50) and GPT-5.6 Sol ($4 / $0.40 / $20) match the rates already in this file.
• A search summary of OpenAI's launch post gives the same three figures.
• Cache write is not in that table. $2.50 follows the 1.25x-input pattern of the other OpenAI rows and matches third-party listings.
• OpenAI's own page returned 403 to the fetch, so this is corroborated from secondary sources, not read from the primary one. The comment in the file says so.

Costs in reports are re-priced from token counts, so a corrected rate later only needs this entry changed.

Testing

$ uv run harness-test validate --static-only   → Static validation passed.
$ uv run pytest -q                             → 766 passed

🤖 Generated with Claude Code