The AI coding cost & usage library
Straight answers on what your AI coding agents cost and how close you are to your limits — model pricing, calculators, rate-limit explainers, and honest comparisons. Claude Code, Codex, Grok, Muse Spark, OpenCode Zen, and OpenCode Go are tracked today; 59 more providers are on the roadmap.
Pricing accuracy
Local model IDs stay in the total
Coding agents log some models under their own IDs — Grok Build writes grok-4.7-build for Grok 4.7. JP maps 34 local and legacy IDs to the published rate they bill at, so those sessions are counted instead of omitted as unpriced.
Value benchmark
Best bang for the buck
Ranked from JP the Cat's bundled model pricing table using a standard coding-agent session — 500K input, 100K output, and 2000K cache-read tokens — then divided by qualitative effort (high / medium / low); models without an effort label are not ranked. Higher effort at lower session cost wins. Every non-Claude figure is notional list price, not subscription spend.
| # | Model | Effort | Session cost | Basis |
|---|---|---|---|---|
| 1 | Muse Spark 1.3 ContributorMeta (Muse Spark) | High | $0.07 | Notional |
| 2 | GPT-5.6 LunaOpenAI (GPT / Codex) | High | $0.26 | Notional |
| 3 | GLM 5.3 FlashOpenCode Zen / Go models | Medium | $0.19 | Notional |
| 4 | MiMo V2.5OpenCode Zen / Go models | Low | $0.10 | Notional |
| 5 | Gemini 2.5 Flash-LiteGoogle Gemini | Low | $0.11 | Notional |
| 6 | GPT-6 LunaOpenAI (GPT / Codex) | Low | $0.12 | Notional |
| 7 | LongCat 2.0OpenCode Zen / Go models | Medium | $0.28 | Notional |
| 8 | Qwen 3.8 FlashOpenCode Zen / Go models | Low | $0.15 | Notional |
See blended $/MTok or tokens-to-dollars to model your own mix.