price fable 5.1, opus 5.5, sonnet 5, gpt-6 and gpt-5.6 at current rates

[pedroapfilho]

Prices were matched by family prefix, so newer models were billed at the previous generation's rates, and gpt-6 fell through to the default Sonnet price.

Pricing (src/pricing.ts)

Per million tokens (input / output / cache read / cache write):

Model	Before	After
claude-fable-5-1	10 / 50 / 1 / 12.5	10 / 50 / 0.25 / 12.5
claude-opus-5-5	5 / 25 / 0.5 / 6.25	4 / 20 / 0.2 / 5
claude-sonnet-5	3 / 15 / 0.3 / 3.75	2 / 10 / 0.2 / 2.5
gpt-6-astra	3 / 15 (default)	10 / 50 / 1 / 12.5
gpt-6-sol	3 / 15 (default)	2 / 10 / 0.2 / 2.5
gpt-6-luna	3 / 15 (default)	0.1 / 0.5 / 0.01 / 0.125
gpt-5.6-sol	1.25 / 10	4 / 20 / 0.4 / 5
gpt-5.6-terra	1.25 / 10	2 / 12 / 0.2 / 2.5
gpt-5.6-luna	1.25 / 10	0.2 / 1.2 / 0.02 / 0.25
gpt-5.5	1.25 / 10	5 / 30 / 0.5 / 0
gpt-5.3-codex	1.25 / 10	1.75 / 14 / 0.175 / 0

OpenAI rates are the short-context Standard tier from https://developers.openai.com/api/docs/pricing. gpt-5.6-sol's price is a promotion running at least through 2026-11-21. Sonnet 5 cache prices use the standard 0.1x / 1.25x multipliers.

Usage windows

• Claude no longer has separate Opus or Sonnet weekly limits, so the seven_day_opus / seven_day_sonnet fallbacks are removed. Fable's 50% cap on Max and Team premium still arrives as a weekly_scoped limit and keeps its own window.
• Model-scoped windows are ranked by id (weekly_scoped:*, codex_*) instead of by matching model names, which dropped Spark after its removal from Codex.

pi

• The pi providers offer claude-fable-5-1, claude-opus-5-5, claude-sonnet-5 and gpt-6-astra, gpt-6-sol, gpt-6-luna. The ChatGPT codex backend now lists the whole GPT-6 family for subscription accounts, so the single-id restriction is gone.
• The TUI demo fixtures use current model ids.

bun run check passes (typecheck, biome, 90 tests).