One API key, every current Claude model. Same context windows, same SDKs, 90% below official pricing. New models are added within days of release.
Every card below is one model id from the models API (GET /v1/models) — the exact string you pass as model to the Messages API. The large language models listed are developed by Anthropic and served here unchanged; the ids match the Claude platform docs, so code written against them works on either endpoint. Family names (Opus, Sonnet, Haiku, Fable) are tiers; the version suffix is what the price is attached to. GPT models (GPT-6 Astra / Sol / Luna, the GPT-5.x line) are served through the Codex group on the OpenAI-compatible endpoint — create a key in that group to use them; gpt-image models are also available there with per-image billing.
Anthropic's newest flagship: the Fable line's strongest coding and agentic model, with a 1M-token context for whole-repository and multi-document work.
Anthropic's current default for long-running agentic coding and knowledge work: Opus-class reasoning at a lower price than Opus 5 ($4 / $20 per million), 1M context, adaptive thinking always on.
The 5.5-generation Sonnet: the best combination of speed and intelligence in the Claude lineup at the Sonnet price, with a 1M context window and adaptive thinking.
The fifth-generation Opus: deep reasoning and long-horizon agent work at 1M context, at the same list price as Opus 4.8. An extended-thinking variant is available.
The everyday production model and the most-called model on this gateway: strong coding at a third of Opus pricing, with a 1M-token context window.
Anthropic's newest 2026 model. Frontier coding and agentic capability with a 1M-token context variant for huge codebases and document sets.
How expensive is the Claude Opus 4 API? Official list price is $5 per million input tokens and $25 per million output; the table below shows the gateway rate. Anthropic's most capable model — deepest reasoning, best-in-class coding and agentic performance, with effort control for cost tuning.
The Opus generation that introduced Anthropic's newer tokenizer and the 1M window at standard pricing. Still served; the same list price as Opus 5.
Opus 4.6: the first Opus with a native 1M-token context at standard pricing and adaptive thinking. Still available for pinned workloads at $5 / $25.
The workhorse. Near-Opus quality on most coding tasks at a fifth of the price — the default choice for production workloads.
The late-2025 Opus that cut the flagship price to $5 / $25. A 200K window and extended thinking; served under its dated id for teams that pinned it.
The Sonnet that made agentic coding mainstream in 2025. Served under its dated id at $3 / $15 for teams that have not moved to Sonnet 5 yet.
Small, fast, and shockingly capable — classification, extraction, routing, and high-volume tasks at the lowest Claude price point.
The 1M-token context tier — load entire repositories, hundreds of documents, or week-long agent transcripts into a single request.
OpenAI's top GPT-6 tier: the deepest reasoning in the family, with effort levels up to max. Served through the Codex group at $10 / $50 per million.
The mid-tier GPT-6: most of Astra's capability at a fifth of the price ($2 / $10). The sensible default for Codex and OpenAI-SDK apps on this gateway.
The small, fast GPT-6 at $0.10 / $0.50 per million — classification, extraction, routing and high-volume pipelines on the OpenAI-compatible endpoint.
The base GPT-5.6 model at $4 / $20: the generation before GPT-6, still served for teams that built on it.
GPT-5.6 Sol — the reasoning-heavy 5.6 variant at $5 / $30, one of the most used OpenAI models on this gateway through Codex.
GPT-5.6 Terra: the balanced 5.6 tier at $2 / $12 — strong coding at a mid price.
The small GPT-5.6 at $0.20 / $1.20 per million for high-volume, low-latency work.
OpenAI's flagship reasoning model — strong writing, math, and multimodal performance through the same OpenAI-compatible endpoint.
GPT-5.4 (March 2026) at $2.50 / $15 with a 1.05M context — served under both the plain and the dated id (gpt-5.4-2026-03-05).
The compact GPT-5.4 at $0.75 / $4.50 — fast and cheap for volume work on the OpenAI-compatible endpoint.
The Codex-tuned 5.3 model at $1.75 / $14, served on the Responses API path for Codex CLI and agent tooling.
GPT-5.2 (December 2025) at $1.75 / $14 — the 5.2 base model, also served as gpt-5.2-2025-12-11 and gpt-5.2-chat-latest.
The Pro tier of GPT-5.2: maximum-compute reasoning at $21 / $168 per million on the Responses API. For the few tasks that justify it.
Google's frontier multimodal model — native 1M context, strong video/image understanding, and top-tier benchmark scores.
Google's speed-tier model — 1M context at bargain rates, ideal for high-volume multimodal and retrieval workloads.
One of 2026's strongest open-weight frontier models — competitive coding and agentic scores at a fraction of closed-model prices.