Claude Opus 5 tops the intelligence index at lower cost than its closest rival

Anthropic's Opus 5 scored 61 on Artificial Analysis's index while running cheaper per task than Fable 5 — but its hallucination rate climbed.

Anthropic released Claude Opus 5, and Artificial Analysis scored it 61 on its Intelligence Index, a composite of nine tests. That places it just above Claude Fable 5 at 60, GPT-5.6 Sol at 59, and its own predecessor Opus 4.8 at 56. The testing was done in collaboration with Anthropic ahead of public release, so the numbers deserve the usual caution.

The cost picture is the notable part. An average Intelligence Index task ran $2.03 on Opus 5 versus $2.75 on Fable 5, with token pricing at $5 per million input and $25 per million output. On the harder AA-Briefcase benchmark, Opus 5 reached 1720 Elo at max reasoning — 146 points above Fable 5 — while cost per task dropped about 20 percent to $17.79. On coding it shares first place on the Coding Index and matched Sol at 89 percent on Terminal-Bench v2.1.

The headline caveat: Opus 5's hallucination rate rose 14 points to 50 percent, and Epoch AI's capability score of 159 actually trailed Fable 5's 161. Vals.ai also notes the highest reasoning tiers produce more complex, error-prone solutions, with 'max' needing over 36 minutes per task.

Why it matters: frontier-grade reasoning getting cheaper per task keeps the cost of machine cognition bending downward, widening who can afford it. Watch whether the elevated hallucination rate limits use in high-stakes settings, and how independent post-release testing shifts these numbers.

Source: The Decoder