GPT-6 Astra ties Claude Fable 5.1, costs less
Independent benchmarks from Artificial Analysis put GPT-6 Astra level with Claude Fable 5.1 on both of the firm's flagship indices, at roughly 40–60% of Fable's cost per task.
As we reported at launch, OpenAI has released GPT-6 Astra. Independent results are now in. Artificial Analysis found that the model ties Claude Fable 5.1 for first place on both its Intelligence Index and its Coding Agent Index.
On the Intelligence Index, Astra at max effort scores 53, the same as Fable 5.1, and 6 points above GPT-5.6 Sol. Each task costs $3.26, compared with $7.63 for Fable. Astra also uses far fewer tokens: about 27k output tokens per task, against 78k for Fable. Every reasoning effort level sits on the cost frontier, from $0.82 per task at low effort to $3.26 at max.
On the Coding Agent Index, Astra running in Codex scores 62. That matches Fable 5.1 in Claude Code and beats Claude Opus 5 (60) and GPT-5.6 Sol (55). At $7.09 per task, Astra is about 40% cheaper than Fable.
Per-token prices are higher. OpenAI charges $10 per million input tokens and $50 per million output tokens, 2.5x GPT-5.6 Sol's $4/$20. Because Astra uses fewer tokens, the cost per task rises less than the price per token.
The biggest gain is in hallucination. On AA-Omniscience, Astra's hallucination rate drops from 92% to 51%, and its accuracy rises by 4 points. It leads three benchmarks: Terminal-Bench v4.0 at 59%, AutomationBench-AA at 69%, and GDP.pdf at 31%.
Some results fell. Astra scores about 45 Elo points below Sol on GDPval-AA v2. The firm noted that Astra used far fewer turns per task: 24, against 45 for Sol and 60 for Fable. Astra also scores lower than Sol on Presentation Quality in AA-Briefcase, though its overall Elo there is about 90 points higher.