Claude Fable 5 Launch: Anthropic’s First Public Mythos-Class Model
·New model claude-fable-5 (released by Anthropic on 2026-06-09, the first publicly available Mythos-class model) — 1M-token context, 128K max output, text/image/file input, callable directly via the Anthropic protocol at /v1/messages
·Priced at 30% of the official API price: $3 input / $15 output per million tokens; cache read $0.30, cache write $3.75 — cache tokens are still itemized separately, never double-charged
·No configuration changes needed for existing API keys — just switch the model field in your requests to claude-fable-5; the Models and Pricing pages are already updated
·claude-opus-4-8 is on sale as well (the official fallback target for Fable 5 high-risk requests) — live price on the pricing page
v1.1.2Improvement
Pricing Page Upgrade: Sticky Toolbar, More Sort Modes, Clearer Model Protocols
·Pricing page top toolbar is now sticky during scroll — search, provider filter, API-format filter, and sort dropdown stay reachable as you browse the list
·Three sort modes: by provider (default), by input price ↑, by output price ↑ — easier side-by-side price comparison
·The "OpenAI" / "Claude" compatibility tags on each model row are now backed by explicit per-model configuration instead of provider-name heuristics — what you see is what actually works
·Protocol-matching upgraded: using the wrong SDK (e.g. an Anthropic-format request for an OpenAI-only model) now returns a clearer error sooner — no more digging through logs to figure out who 502'd
·Polish: the sticky toolbar uses a translucent + frosted-glass background in both light and dark themes so it never obscures the row beneath
v1.1.1Improvement
Image API: 2048×2048 and More Sizes Now Supported
·Image endpoints (`/v1/images/generations` and `/v1/images/edits`) drop the 8-size enumerated whitelist in favour of transparent forwarding — any size the upstream model accepts now works
·Verified-working new sizes: `2048x2048` (high-res square), `1280x1280` (non-standard square), `1024x1792` and `1792x1024` (7:4 tall / wide)
·Lightweight format guards remain (must be `auto` or a `WIDTHxHEIGHT` string, ≤16 characters) so malformed inputs do not burn resources
·Upstream-rejected extreme sizes (e.g. `4096x4096`) come back as a 502 with the pre-deduction automatically refunded — no charge to your balance
·API docs "Supported Sizes" section updated with a verified-working table
·Internal byte-level audit confirmed: source upload to R2 and upstream output persistence to R2 do not re-compress — SHA256 matches end-to-end
v1.1.0Feature
Image Generation Now Available
·New Image Generation page powered by the gpt-image model family — describe an image in plain language and generate it instantly
·Full parameter control: size, quality, background transparency, and batch count, with "Auto" as the sensible default
·Conversational interface redesign: the composer is pinned to the bottom of the page, parameters are collapsed into chip-style pills above it, Enter to send and Shift+Enter for a new line — IME-friendly for Chinese input
·Unified history view with single-image and batch (ZIP) download, plus click-to-zoom preview
·Per-image billing for image models, with cost breakdown shown directly under the result and a dedicated "Per Generation" column on the Pricing page
·Content-policy detection upgraded: disallowed prompts surface as a friendly inline notice and never consume balance
v1.0.0Feature
Initial Release
·Tokensmart launched with a unified API gateway for OpenAI / Claude / Gemini / DeepSeek and more
·Pay-as-you-go pricing with a single API key for all models
·Complete dashboard for usage, billing, logs, and API key management