AI model list prices and how to read a rank

Vendors bill input and output separately, per million tokens. Input is what you send, including the instructions and the retrieved notes. Output is what the model writes back, including the reasoning some models spend before the answer. A loop multiplies both, because every extra step is another call.

The table above is the standard list price on each company’s own API, for a prompt inside that vendor’s short-context band, read on November 1, 2026. Cached input is cheaper. Batch is often half. A very long prompt can reprice the whole request. A cloud reseller can charge a different number. Use the linked page before you budget.

A lab can be first on a coding test and ordinary on a fraud test. The honest sentence names the test, the date, and who measured it.

On October 27, 2026, Artificial Analysis replaced its Intelligence Index. Sixty percent of the new index uses private tests or held-out answers. Artificial Analysis put Claude Opus 5.5 first at 58. On October 28, Google DeepMind said Gemini 4 Carbon scores 59 on that same index. That 59 is DeepMind’s report; Artificial Analysis’s announcement did not list Carbon, and Carbon is not on the public API yet.

An open-weight release is the same habit with a file attached. On October 29, 2026, Mistral released weights for Large 4 (Le Chonk): about 1 trillion parameters, roughly 49 billion active, a 1 million token context. Mistral reports 41 percent on Terminal-Bench 4.0 and 48 percent on the Artificial Analysis Coding Agent Index, with 47 percent fewer tokens than GLM-5.3 on that index. Those percentages are Mistral’s citations. The downloadable checkpoint is a separate fact from any one leaderboard row.

The rank that changes a team’s work is a held-out set of the actual job. Coinbase did not ask which model was famous. It asked which system caught Onramp fraud, on four detection metrics, at what delay. On that private test, a post-trained 9B model beat Claude Opus 4.5. The percentages are Coinbase’s. The test is not public, so another company cannot rerun it. Another company can copy the method.

Not every important number is a leaderboard. Epoch’s October 29 report says open-weight GLM 5.3 compromised six recent Firefox releases by reading security fixes that were public but not yet shipped, at about $1,000 of API spend per version after the harness existed. OpenAI’s October 29 monitor note prints a 30-minute alert and a more-than-30-hour lead on a Hugging Face incident replay. Those are measurements: who ran them, on what setup, and what they do not prove.

Read a security or agent-capability post the same way you read a rank. Name the lab or paper, the date, and the concrete figure. Ask whether the number is a live case, a replay, or a lab harness. Epoch has not confirmed the same Firefox attack on version 159.0. OpenAI’s 30-hour lead is OpenAI’s simulation of its own past incident. A dollar cost after a $100,000 setup is not the same as a free exploit for everyone.