Use cases
Which models fit coding agents, code review, batch jobs and large codebases, and what each choice costs per million tokens.
- Running Claude Code against token-shareClaude Code speaks the Anthropic Messages API, which this gateway serves at /v1/messages. Point the base URL at it, pin the tier variables, and set the context to the id you chose.
- Claude Haiku 4.5 for cheap tasks: check the multiplier firstThe only Haiku id here is a 0.4× stable lane at $0.40/$2.00 per million — cheaper than every other dedicated Claude lane, but 1.33× claude-sonnet-5 on the pool lane.
- Running Codex CLI against token-share.appCodex CLI speaks the responses wire format, which is what the OpenAI ids here answer on. One custom model_provider block points it at the gateway — and one id in the family will not work this way.
- Gemini 3 Flash: the lowest per-token price in the catalog$0.050 per million input and $0.30 per million output after the 0.1x multiplier, with a 1,000,000-token context. What that buys, what the runners-up cost, and when the cheapest id is the wrong one.
- GPT 5.4 Mini: the budget OpenAI id and its ceilingsAt 0.1× on a $0.75 / $4.50 list rate, gpt-5.4-mini works out to $0.075 / $0.45 per million tokens. It keeps tools and thinking; what it caps is context and output. And one sibling undercuts it.
- Is grok-composer-2.5-fast cheap? What 0.2x of $3/$15 actually costsEvery Grok id is 0.2x, but on a $3/$15 official rate that lands Composer at $0.60/$3.00 per million — 1.5 times the other Grok text ids on input, 2.5 times on output, and four times gemini-3.8-flash at the same multiplier. Plus no thinking and a 200,000-token window.
- Models for agents: tool calling, context growth and the billEvery text model here supports tool calling. What separates them for a 40-turn loop is context headroom and input price — the transcript is re-billed on every turn.
- Models for batch processing: when the bill is the requirementClassify 100,000 documents and the same job runs $7.60 on Gemini 3 Flash, $56 on Claude Haiku 4.5 Stable and $84 on Grok Composer 2.5 Fast. The multiplier is not the price.
- Models for code review: what a diff review actually costsReviewing a diff is input-heavy and output-light, so the input rate decides the bill. The same 100k-token review runs $0.033 on Claude Sonnet 5 and $0.0056 on Gemini 3 Flash.
- Models for large codebases: when context size is the deciding factorFourteen ids here take 1,000,000 tokens and GPT 6 Astra Stable takes 1,050,000. Filling a 1M window costs $0.050 on Gemini 3 Flash and $2.00 on Astra — per request, every request.