Anthropic · updated 2026-08-29
Run Claude Sonnet 5: providers, pricing and flat-rate access
Anthropic's mid-tier Claude 5 model — the default workhorse for Claude Code and agentic coding, balancing frontier-adjacent quality with mid-tier pricing.
Three real routes to Sonnet 5: Claude Pro/Max plans ($20–200/mo, metered by session windows and weekly caps), the Anthropic API per-token ($2/$10 per M tokens), or a flat-rate provider speaking the Anthropic Messages API — Standard Compute serves Sonnet 5 over /v1/messages at a fixed monthly price, so Claude Code works via a base-URL swap with no windows.
Your options, honestly
Claude Pro / Max$20–200/moFirst-party plans with the deepest Claude Code integration — but metered by 5-hour session windows plus weekly caps that heavy users hit.
Anthropic APIPay-per-token$2/$10 per M tokens (the introductory rate Anthropic kept as standard); no windows, but agentic context growth compounds the bill.
Standard Compute$39+/mo flatSonnet 5 over /v1/messages and in the OpenAI-compatible routed pool — fixed price, no session windows. Routing decides when Sonnet-class is used; no pinning.
Bedrock / VertexPay-per-tokenEnterprise route under cloud billing; same per-token economics.
the flat-rate trade-off
The honest trade-off: first-party plans give you every Anthropic variant and native integration; the flat-rate route gives you Sonnet 5 quality without windows, weekly caps, or a meter. Flat price, not infinite compute — but a heavy Claude Code week costs the same as a quiet one.
Running it in your agent
Claude Code: set ANTHROPIC_BASE_URL=https://api.stdcmpt.com and run — no session windows. OpenAI-compatible agents (OpenCode, Cline, Aider, Roo) reach Sonnet-class through the routed 'standardcompute' model.
FAQ
Why do I keep hitting limits on Sonnet 5 with a Claude plan?
Pro/Max plans meter by 5-hour session windows plus weekly caps, and agentic sessions burn windows fast. The alternatives are per-token API billing (no windows, unbounded cost) or a flat-rate Messages-API provider (no windows, fixed cost).
Is Sonnet 5 enough, or do I need Opus/Fable?
For most agentic coding, Sonnet 5 is the sweet spot — that's why it's the Claude Code default. The top tiers earn their price on the hardest reasoning steps, which is exactly the split routing makes per-request in a mixed pool.
More: how flat-rate LLM APIs work · per-agent guides for OpenCode, Claude Code and Cline · model comparisons