OpenAI Swaps GPT-6.1 Sol In for GPT-6 Sol Just a Week After Launch, Nearly Matching Flagship at a Quarter of the Cost

OpenAI has replaced GPT-6 Sol with a new model, GPT-6.1 Sol, just seven days after the original's release, according to benchmarking data reported via Hacker News. The rapid turnaround underscores how quickly frontier AI labs are now iterating on mid-tier models.

On Artificial Analysis's Intelligence Index, GPT-6.1 Sol scores just one point below OpenAI's flagship model, GPT-6 Astra, while costing less than a quarter as much per task. The new model gained four points over GPT-6 Sol and five points over GPT-5.6 Sol on the index, with particularly strong improvements in agentic knowledge work: it rose four points on AA-Briefcase v1.1 and five points on GDPval-AA v2.1. Other gains include a 12-point jump on Terminal-Bench 4.0, five points on Humanity's Last Exam, six points on GDP.pdf, and eight points on AA-Omniscience Accuracy, alongside a drop in hallucination rate from 60% to 54%.

Pricing remains unchanged from GPT-6 Sol at $2 per million input tokens and $10 per million output tokens, but the cache read discount increases from 90% to 95%, giving GPT-6.1 Sol a slightly lower blended cost for agentic workloads. That follows GPT-6 Sol's original 50% price cut relative to GPT-5.6 Sol, extending a broader trend of falling costs across OpenAI's model lineup.

At maximum effort, GPT-6.1 Sol costs $0.72 per Intelligence Index task, compared with $3.26 for GPT-6 Astra, $1.05 for GPT-6 Sol, and $1.99 for GPT-5.6 Sol. Every effort tier of the new model pushes out the cost-efficiency Pareto frontier, meaning no cheaper model currently matches its intelligence level for a given price.

Token efficiency is more mixed: GPT-6.1 Sol uses roughly 10% to 30% more output tokens than GPT-6 Sol across effort settings. Still, because of its higher intelligence scores, the model's low and medium effort tiers remain Pareto-optimal for token efficiency.

In coding tasks, GPT-6.1 Sol gains three points over GPT-6 Sol at maximum effort on the Artificial Analysis Coding Agent Index, landing two points below GPT-6 Astra. At its highest effort setting, labeled "xhigh," the model scores one point above GPT-6 Astra while costing less than 15% as much per task — a six-point gain over GPT-6 Sol's peak performance. Artificial Analysis noted that the xhigh setting outperformed the standard max effort setting by three points, and that GPT-6.1 Sol dominates the lower-price end of the Pareto frontier for the Coding Agent Index.

On AA-Omniscience, GPT-6.1 Sol improved eight points in accuracy at maximum effort while cutting its hallucination rate by six points. On AA-Briefcase, the model gained roughly 80 Elo points, driven by improvements in rubric scoring and Analytical Quality Elo, even as its Presentation Elo dipped slightly.

The figures come from Artificial Analysis's Intelligence Index v4.3.2 benchmark suite. As with any third-party benchmarking report, the underlying methodology and full evaluation breakdowns should be reviewed directly on Artificial Analysis's website before figures are cited elsewhere.