GPT-6.1 Sol replaces GPT-6 Sol after just 7 days, with near-Astra intelligence

GPT-6.1 Sol replaces GPT-6 Sol after seven days, delivering near-Astra intelligence while cutting costs. It scores one point below GPT-6 Astra on the Intelligence Index, up 4 points versus GPT-6 Sol and 5 versus GPT-5.6 Sol. Pricing retains GPT-6 Sol’s $2/$10 per million input/output tokens but raises the cache-read discount from 90% to 95%, yielding a slightly lower blended price for agentic workloads. At max effort, cost per Intelligence Index task is $0.72 versus GPT-6 Astra’s $3.26, 31% cheaper than GPT-6 Sol ($1.05) and 64% cheaper than GPT-5.6 Sol ($1.99), and every effort level expands the cost-efficiency Pareto frontier.

Benchmark detail shows substantive capability gains in agentic knowledge work and coding. AA-Briefcase v1.1 and GDPval-AA v2.1 improve by 4 and 5 points; Terminal-Bench jumps 12 points; Humanity’s Last Exam +5 and GDP.pdf +6. AA-Omniscience accuracy rises 8 points at max effort while hallucination rate falls from 60% to 54%. Token use increases ~10-30% in output tokens, yet low/medium effort settings are Pareto optimal for token efficiency. The Coding Agent Index gains 3 points at max effort versus GPT-6 Sol and sits 2 points below Astra; the xhigh setting can outscore Astra by 1 point for under 15% of Astra’s cost, and AA-Briefcase gains ~80 Elo driven by higher rubric and analytical-quality scores with a slight drop in presentation Elo.