BenchLeader

What your AI use costs on each model

Say what you use it for and roughly how much. Every ranked model prices that month at its list prices, the best one your budget buys comes first, and for coding you see which flat-rate plans would cover it. It all runs in your browser.

248 ranked models with list prices · data as of 15 Nov 2026

What for

Claude Code, Codex, Cursor and the like. Agents re-read their context every turn, so most tokens are cache reads.

How much a month500M tokens
Budget a month$100
Only

Your month: 500M tokens (97% cache reads, 3% fresh input, 1% output).

  • Best you get for $100 a month
    Gemini 4 Carbon

    About $99 a month at API prices. Coding score 74; the top model, Claude Opus 5.5, would cost $197.

  • Nearly as good, for much less

    Nothing within 3 points costs under 60% of the pick, so the pick is also the value choice.

Or a flat-rate plan

Plans whose estimated monthly allowance covers your month on the model, cheaper than paying the API for it.

  • Claude Code Profor Claude Opus 5.5$20 instead of $197
  • Codex Pro 200for GPT-6 Astra$200 instead of $735
  • Claude Code Profor Claude Sonnet 5.5$20 instead of $99
  • Codex Plusfor GPT-6.1 Sol$20 instead of $99
  • Codex Plusfor GPT-6 Sol$20 instead of $147

Allowances are estimates and plans have weekly limits on top. How each plan compares.

Your month on each model

ModelCodingA monthSpeed
Claude Opus 5.580$197 (over budget)114 tok/s
Claude Fable 5.280$297 (over budget)102 tok/s
Gemini 4 Carbon74$9961 tok/s
GPT-6 Astra73$735 (over budget)40 tok/s
Claude Sonnet 5.572$99143 tok/s
GPT-6.1 Sol71$9949 tok/s
Claude Fable 5.171$371 (over budget)88 tok/s
Claude Fable 571$735 (over budget)50 tok/s
GPT-6.1 Terra69$4971 tok/s
Gemini 4 Hydrogen69$74169 tok/s
Kimi K3.1 Pro68$147 (over budget)30 tok/s
Claude Opus 568$368 (over budget)80 tok/s
o3-prono cache price67$10,150 (over budget)–
Gemini 4 Argon67$9963 tok/s
GPT-6 Sol66$147 (over budget)117 tok/s

Greyed rows are over your budget. List prices as of 15 Nov 2026; a provider’s own discounts are not included.

Questions people ask

How do I know how many tokens I use?
Coding agents report it: Claude Code shows usage per session, Codex and Cursor show it in their usage pages, and an API dashboard (OpenAI, Anthropic, OpenRouter) shows a month's input and output tokens. As a rough guide, a developer running an agent most workdays reads about half a billion tokens a month, nearly all of it from the prompt cache.
Why are coding months so many tokens?
An agent sends its whole context (files, tool results, the conversation so far) on every turn. Providers cache that prefix and charge a fraction of the input price to read it, which is why the calculator prices a coding month at 97% cache reads, the same mix the coding plans page uses.
Which score ranks the models?
For coding, the coding category of the BenchLeader Index; for chat, the Index itself; for your own app, the skill you pick. Scores are on the site's scale, where 50 is the average evaluated model and 15 points is one standard deviation.
Are the plan allowances exact?
No. Most vendors do not publish their limits in tokens, so the allowances are Real API Pricing's estimates from measured use, and plans add weekly caps on top. GitHub Copilot is the exception: it publishes its allowance in dollars of API usage.

Related: coding plans compared, find a model by constraints, cheapest models, LLM price statistics, best cheap LLMs.