feat(pricing): add Fable 5.1 and Mythos 5.1

[SergeAx]

Adds the two rows Anthropic published for Claude Fable 5.1 and Claude Mythos 5.1, and bumps pricing_version to 2026-09-02.1 (rule 8).

The one thing worth reading twice

Both models cost the same as Fable 5 on input and output — $10/$50 per 1M, with $12.50/$20 cache writes — but their cache reads are 0.025x input ($0.25/MTok), not the 0.1x every other row in the table follows. Anthropic's pricing page carries it as an explicit footnote:

Cache hits and refreshes on Claude Fable 5.1 and Claude Mythos 5.1 are priced at 0.025x the base input price. All other models use the standard 0.1x multiplier.

The table's own notes previously stated the 0.1x rule without exception, so anyone deriving cache_read from it would have written $1.00 and overcharged a cached Fable 5.1 turn four times over. The note now records the exception and why those rows break the rule.

TestFable51CacheReadsAtItsOwnRate pins it. claude-fable-5-1 is one character away from claude-fable-5, whose cache reads cost 4x as much; longest-prefix-wins already resolves it correctly, and the test makes that a guarantee rather than an accident.

Mythos 5.1 is included because the table already pairs every Fable row with its Mythos twin, and the pricing page lists both at identical rates.

Provenance

All figures read from https://platform.claude.com/docs/en/about-claude/pricing on 2026-09-02. Top-level fetched_at is deliberately left at 2026-08-18 — see the open issue below for why bumping it would have asserted more than this PR delivers.

Checks

go vet, go test ./..., UI tests, lint-ui, dist-check and make smoke are green. Two caveats, neither from this change:

• internal/opencode TestIngestLive fails on master too (it reads a live OpenCode database: sql: Scan error on column index 1, name "c": converting NULL to float64 is unsupported).
• golangci-lint is not installed on this machine, so lint-go's second half ran in CI only.

Out of scope, but found while sourcing this — Sonnet 5 is now overcharged

The same pricing page says:

The $2/$10 per million input/output token pricing for Claude Sonnet 5, announced at launch as introductory pricing through August 31, 2026, is now the standard price. The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur.

pricing.json still has the until: "2026-08-30" row at $2/$10 and a current row at $3/$15, so every Sonnet 5 turn priced since 2026-08-31 is reported 50% high. That is a rule 6 problem in the opposite direction from the one dated rows were built to prevent. It wants its own commit — happy to open it — but it did not belong in a PR titled "add Fable 5.1".

https://claude.ai/code/session_01PbAUomrYzs8MeJbd66Q1QP

[dspv]

Thanks — merged.

The cache-read test is the part I'd have wanted most: $0.25 sits one character away from a row that charges four times as much, and it's the first row to break the 0.1× rule.

Verified the figures against the pricing page, and checked that claude-fable-5-1 doesn't collide with claude-fable-5 — longest-id-first handles it.

I'll fix fetched_at on merge; it was stale before this PR.

First outside contribution here, and it came with a test that would catch a real mistake. Appreciated.