OpenAI drops GPT-6.1 Sol one week after GPT-6 Sol, nearly matches Astra at a quarter the cost

OpenAI released GPT-6.1 Sol just seven days after GPT-6 Sol. The new model sits one point below GPT-6 Astra on the Intelligence Index while costing less than a quarter per task. List pricing is unchanged at $2 per million input tokens and $10 per million output tokens (roughly 7.4 and 37 shekels), but the cache-read discount rose from 90% to 95%, shaving a bit more off the blended price for agentic workloads. That follows a 50% discount GPT-6 Sol already offered versus GPT-5.6 Sol.

Performance jump over the previous generation

Artificial Analysis data shows GPT-6.1 Sol improves 4 points on the Intelligence Index over GPT-6 Sol and 5 points over GPT-5.6 Sol, landing one point shy of Astra. Knowledge-work benchmarks rose 4 points on AA-Briefcase v1.1 and 5 on GDPval-AA v2.1. Other notable gains: 12 points on Terminal-Bench 4.0, 5 on Humanity's Last Exam, 6 on GDP.pdf, and 8 on AA-Omniscience accuracy. Hallucination rate fell from 60% to 54%.

Record cost efficiency

At max effort, GPT-6.1 Sol costs $0.72 per Intelligence Index task versus Astra's $3.26 — a more than 4× gap. Against GPT-6 Sol that's a 31% saving ($1.05 per task), and against GPT-5.6 Sol a 64% drop (from $1.99). Every effort tier of the new model pushes the cost-efficiency Pareto frontier: for any given intelligence level, no model is cheaper.

Token efficiency: more value despite more output

The new model emits 10–30% more output tokens than its predecessor at every effort tier. Yet because intelligence scores rose, the low and medium tiers become Pareto optimal for token efficiency — more intelligence per paid token.

Coding gains, too: narrowing the gap with Astra

On Artificial Analysis's Coding Agent Index, GPT-6.1 Sol adds 3 points over GPT-6 Sol at max effort and sits 2 points below Astra. The improvement extends a consistent trend: smaller, cheaper models are closing the gap on flagships, making Astra a less obvious choice for most workloads.