Skip to content

v1.22.1 — Corrected vendor model IDs and prices

Choose a tag to compare

View all tags
@stackbilt-admin stackbilt-admin released this 15 Oct 20:04
v1.22.1
da86fb3
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Fixed

  • Vendor model IDs and prices (#112, phase 1), checked against primary vendor docs on 2026-10-15:
    • Anthropic 4.6 models use their real dateless IDs (claude-opus-4-6, claude-sonnet-4-6), and Claude Opus 4 uses claude-opus-4-20250514. The invalid -20250618 IDs earlier releases published are still accepted: adapters, catalog lookups, routing preflight, gateway plans, the factory (quota checks, hooks, cache keys) and CreditLedger spend and per-model budgets all translate them to the real ID. MODELS.CLAUDE_OPUS_4_6, CLAUDE_SONNET_4_6 and CLAUDE_OPUS_4 now hold the real IDs. Because MODELS is as const, their literal TypeScript types change; code that pins the old literal strings, or compares persisted values against MODELS.*, needs updating. The old values never worked against the Anthropic API.
    • Claude Opus 4.6 is priced at $5/$25 per million tokens (was $15/$75). Opus 4.6 and Sonnet 4.6 report their 1M-token context window (was 200K). Claude 3.5 Haiku is priced at $0.80/$4 (was $0.25/$1.25).
    • Anthropic usage cost now includes prompt-cache reads (0.1x input) and writes (1.25x for 5-minute, 2x for 1-hour). Streaming usage takes the latest cumulative counts from message_delta as well as message_start.
    • Claude models retired on the Claude API (Opus 4, Sonnet 4, 3.7 Sonnet, 3.5 Sonnet/Haiku, 3 Opus/Sonnet) are marked retired and no longer chosen by routing. Explicit requests still pass through.
    • Workers AI prices for Llama 3.1 8B and Llama 3 8B ($0.282/$0.827 per million) and Mistral 7B v0.1 ($0.110/$0.190) replace placeholders that were up to 8,270x too low. Workers AI models not on Cloudflare's pricing page are labelled unverified.
    • gpt-4, gpt-4-turbo and gpt-3.5-turbo carry OpenAI's 2026-10-23 shutdown date. The unsupported "gpt-4o retired 2026-04-03" claim is removed.
    • Cerebras prices are labelled unverified placeholders, since Cerebras publishes no per-token prices.

Pull requests

  • fix: correct vendor model IDs and prices against primary docs (v1.22.1) (#122)

Full Changelog: v1.22.0...v1.22.1

Assets 2
Loading