How much AI can you afford?

How to read these numbers

• Tokens per $1 — how many tokens a dollar buys at each model's blended input/output price. This is the raw, objective part of the score: a model that's cheap on output (which costs ~3× input) ranks higher here.
• Intelligence per $1 — tokens per dollar multiplied by each model's intelligence index. It stops the metric from simply crowning the cheapest model, and it's the ranking the calculator sorts by.
• Intelligence index — a 0–100 benchmark of general capability (from Artificial Analysis). 60 means roughly 1.5× the capability per token of a model at 40.
• Last verified — every figure shows the date it was checked. AI prices move weekly, so we never pretend a number is "current" longer than it is.

The full formula, its limits, and our data sources are documented on the methodology page — we'd rather you judge the metric than trust it.

Top picks right now

What changed recently

• 19 Sep 2026 — Step 5 Preview added — a new lab at the top of the value table. StepFun's Step 5 Preview joins the catalogue with an index of 44 at $1 / $2.70 per 1M tokens — above Grok 4.6 and Kimi K3, for roughly a tenth of the premium flagships' price. It takes text and image input over a 1M-token context and runs at about 100 tokens per second. Two caveats: it is a preview release, and StepFun publishes no first-party USD price list we could reach, so the price comes from Artificial Analysis. It carries a review flag until we can confirm it.
• 19 Sep 2026 — GLM-5.3 FlashX added. Z.ai's high-speed variant of GLM-5.3 Flash is in the catalogue at $0.37 / $1.25 per 1M tokens with a 1M-token context window and roughly 200 tokens per second — about 2.5x GLM-5.3 Flash's per-token price for a large jump in throughput. Artificial Analysis has not benchmarked it yet, so it carries no Intelligence Index and its rating is a tier estimate rather than a measurement.
• 19 Sep 2026 — Four superseded models retired. Claude Fable 5, Grok 4.5, GLM-5.2 and Gemini 3.6 Flash are now marked obsolete, as each has been replaced by a newer model from the same provider that matches or beats it: Fable 5.1 (53 vs 50 at the same $10 / $50), Grok 4.6 (44 vs 39 at the same $2 / $6), GLM-5.3 (45 vs 34 at the same $1.40 / $4.40) and Gemini 3.8 Flash (41 vs 34 at the same $0.75 / $3.75). Their pages stay online with a link to the successor, and they are hidden from the calculator unless you turn retired models on. Gemini 3.7 Flash remains current — Google still sells it and it is only a month old.

Full history on the What's new page. Prices are refreshed weekly from the OpenRouter catalog and official provider pages.

Go deeper

• Pricing guides — how much ChatGPT, Claude, DeepSeek and Gemini really cost in 2026, and which model to pick for your workload.
• Head-to-head comparisons — DeepSeek vs Claude, Gemini vs ChatGPT, Grok vs Claude and more, with live numbers.
• Subscription vs API — when a flat monthly plan beats metered pricing, and the exact token break-even for each plan.
• The methodology — the exact formula behind every score on this site, and what it does and doesn't measure.