Checked Sep 30, 2026Start a new check

What was checked

GPT-6.1 Sol achieves near-Astra Intelligence (1 point below GPT-6 Astra), with significant gains in agentic knowledge work at less than one quarter of the cost per task

Assessment summary

Accurate

OpenAI’s materials describe GPT-6.1 Sol as delivering near-Astra performance and comparable capabilities to GPT-6 Astra.

This saved result keeps its original verdict. Our current method would assess it differently, so grouped findings are not shown here. Review the model assessments below or start a new check for an updated report.

Model assessments

Assessment summary

  • The claim is partly supported, but one key detail is wrong or too specific.
  • OpenAI says GPT-6.1 Sol delivers near-Astra performance and in DeepSWE v1.1 it matches GPT-6 Astra at roughly one-fifth the cost, but the provided sources do not support the exact phrasing “1 point below GPT-6 Astra” or the broader “significant gains in agentic knowledge work” as a verified benchmark claim.openai.com developers.openai.com

Claims reviewed

GPT-6.1 Sol is 1 point below GPT-6 Astra.

Contradicted
The cited sources instead say it matches Astra on DeepSWE v1.1 or falls within a small margin on other benchmarks, but they do not support a universal claim of being exactly 1 point below Astra.openai.com siliconangle.com vellum.ai

This claim may mislead readers without added context.

GPT-6.1 Sol delivers significant gains in agentic knowledge work.

Supported
OpenAI says it improves complex professional tasks, including multi-step business workflows and computer-use tasks, which fits the general idea of agentic work gains.openai.com developers.openai.com

GPT-6.1 Sol does this at less than one quarter of the cost per task.

Supported
OpenAI reports roughly one-fifth the cost on DeepSWE v1.1, which is less than one quarter.openai.com

Assessment summary

  • The claim is factually accurate across all three components.
  • Multiple authoritative sources confirm that GPT-6.1 Sol scores 1 point below GPT-6 Astra on the Intelligence Index and costs less than one quarter of Astra's cost per task.
  • The specific metrics are well-documented and consistent across sources.

Claims reviewed

Significant gains in agentic knowledge work

Supported
On DeepSWE v1.1, GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth of the cost, while eclipsing GPT-6 Sol's best score by 6.4 percentage points. Multiple benchmarks demonstrate substantial improvements in coding and professional work.

GPT-6.1 Sol achieves near-Astra Intelligence

Supported
GPT-6.1 Sol is described as near-Astra intelligence for coding, computer use, and professional work. This is consistently supported across multiple sources.

GPT-6.1 Sol costs less than one quarter of Astra's cost per task

Supported
GPT-6.1 Sol achieves this at less than one quarter of the Cost per Task. More specifically, at max effort, GPT-6.1 Sol costs less than a quarter of GPT-6 Astra per Intelligence Index task ($0.72 vs $3.26).

Assessment summary

  • Parts of the statement match multiple public announcements and benchmark summaries.
  • However independent verification of the model's real-world "agentic knowledge work" gains is limited and the phrasing mixes vendor claims and benchmark summaries.

Claims reviewed

GPT-6.1 Sol shows significant gains in agentic knowledge work.

Can’t verify
OpenAI and press reports claim improved agentic/problem-solving performance, but independent, peer-reviewed measurements of "significant gains" in broad real-world agentic knowledge tasks are not yet available.

GPT-6.1 Sol operates at less than one quarter of the cost per task compared to GPT-6 Astra.

Supported
Press releases and pricing/benchmark summaries report GPT-6.1 Sol being offered at about one-fifth (≈20%) of Astra's cost per task, which is under one quarter.

GPT-6.1 Sol achieves near-Astra Intelligence (1 point below GPT-6 Astra).

Supported
Public benchmark summaries and third-party comparisons report GPT-6.1 Sol scoring roughly one point lower than GPT-6 Astra on the cited Intelligence/AA indexes.

Sources cited

38 sources

Share this result