Published on [Permalink]
Reading time: 2 minutes

The expensive last two points of AI intelligence

Source: Artificial Analysis, Comparison of Models, viewed 25 July 2026.

The latest Artificial Analysis model comparison puts Claude Opus 5 (max) at the top of its Intelligence Index with a score of 61. GPT-5.6 Sol (max) is close behind at 59.

That is a marginal lead for Anthropic, but it is an expensive one.

Artificial Analysis estimates that a Claude Opus 5 Intelligence Index task costs US$2.03, compared with US$1.04 for GPT-5.6 Sol. In other words, Anthropic’s extra two index points come with a cost premium of about 95 per cent - almost double.

This does not make Claude the wrong choice. If the best available benchmark result matters more than price, Claude is the leader. But for repeated work, GPT-5.6 Sol looks like the more compelling value: about 97 per cent of Claude’s Intelligence Index score for roughly half the task cost.

As always, a benchmark is not the whole experience. Model behaviour, reliability, writing quality and fit for a particular task still matter. Even so, this comparison makes the trade-off unusually easy to see. Anthropic has the highest bar; OpenAI is standing very close to it and charging considerably less.

Artificial Analysis defines cost per task as a weighted average across the evaluations in its Intelligence Index, incorporating input, caching, reasoning and answer-token costs. The figures will change as models and pricing change, so the live comparison remains the best reference.

✍️ Reply by email