In the Intelligence Index, GPT-6 Astra scored 61 points, matching the previous GPT-5.6 Sol. In comparison, Anthropic's Claude Fable 5.1 scored 66 points, and Meta's Muse Spark 1.3 scored 62 points.
The Coding Agent Index results show GPT-6 Astra scored 67 points in the Codex environment. This performance is roughly equal to Claude Opus 5, Fable 5, and Muse Spark 1.3, while Claude Fable 5.1 remains the leader with 70 points.
Pricing for GPT-6 Astra has increased 2.5 times compared to GPT-5.6 Sol. Input tokens now cost $10 per 1 million, up from $4, and output tokens cost $50 per 1 million, up from $20. Cache reading discounts of 90% and writing surcharges of 25% remain unchanged.
The model shows improved token efficiency, generating about one-third of the volume of GPT-5.6 Sol and one-fifth of Claude Opus 5. In coding tasks, this makes the cost comparable to its predecessor, while in the Intelligence Index, it remains 75% more expensive per task than Claude Fable 5.
Testing revealed mixed results: hallucinations dropped from 92% to 51% in the AA-Omniscience test, and the model gained 80 Elo points in the AA-Briefcase test for large projects. However, it lost 80 Elo points in the GDPval-AA v2 professional tasks test.
GPT-6 Astra gained six points in Humanity’s Last Exam but saw slight declines of two to three points in the τ³-Banking, SciCode, and AA-LCR tests.
