AI Report BluNET Studio
Intelligence vs. cost

Model rankings

What the frontier models score on a third-party intelligence index, and what they cost per million tokens. The AI Report updates this when an issue moves a number.

2 index points separate Claude Opus 5 from Grok 4.6
76% less per output token for Grok 4.6 ($6 vs $25 per 1M output tokens)

That is the whole argument the AI Report has been making since late July: base intelligence is converging, and the competition has moved to cost per task, latency, and the harness around the model. A 2-point index gap is not what a 76% price gap usually buys.

The table

Current as of August 24, 2026. Each row shows its own capture date, because these figures come from different weekly issues and were not measured on the same day.

Model intelligence index and token cost, one row per model, each with its own capture date.
Model Intelligence In Out Per task As of What it means
Claude Opus 5 Anthropic 63 $5 $25 · Aug 17 Tops the index. Reported as too expensive to be popular in the following week's coverage.
Claude Fable 5 Anthropic 62 · · · Aug 17 Restricted-access model. No published per-token price we have verified.
GPT-5.6 Sol OpenAI 61 $5 $30 · Aug 17 An Ultrafast tier running the same model up to 14x faster was previewed on 2026-08-13.
Grok 4.6 xAI 61 $2 $6 $0.84 Aug 17 Frontier parity at unchanged pricing - a capability gain, not a price cut. Sits on the index-vs-cost-per-task frontier.
Kimi K3 Moonshot AI open weights 57 $3 $15 · Jul 20 Open weights. Figure is a month older than the rows above it and has not been re-measured here.
Gemini 3.7 Flash Google · · · · Aug 24 Shipped 2026-08-13 with a 50% intro price cut and topped agent benchmarks the following week. We have not seen a published index score or list price to reproduce.

Intelligence figures are the Artificial Analysis Intelligence Index, published by Artificial Analysis and reproduced here with attribution. They are not BluNET Studio measurements. Prices are USD per 1M tokens (input / output). Rows are not a single snapshot: each carries the date it was captured, and a blank means no published figure has been verified rather than zero. Model pricing and scores change without notice.

How to read it

What this table does and does not claim

  • The index is not ours. The intelligence numbers are the Artificial Analysis Intelligence Index, published by Artificial Analysis, a third-party benchmarking organisation. We reproduce them with attribution and we do not run the benchmark ourselves.
  • The rows are not one snapshot. They were captured in different weekly issues, so a row dated July sits next to a row dated August. Each row prints its own date. Comparing two rows with different dates compares two different weeks.
  • A dash means no verified figure, not zero. 1 model in the table has no published index score we have been able to verify. We leave those blank rather than estimating them.
  • List price is not your price. The prices here are published list rates per million tokens. Caching, batch tiers, committed-use discounts and reasoning-token overhead all move real spend, sometimes by more than the gap between two models.
  • A capability gain is not a price cut. The report has corrected itself on exactly this: Grok 4.6 reached frontier parity at unchanged pricing, which reads as a discount only if you compare it to competitors rather than to its own prior price.
Subscribe

Get the movement, not just the table

The weekly issue explains what changed and why it changed. Every Monday.

Free and weekly, with double opt-in. Unsubscribe any time.

Read the latest issue