OpenAI has introduced GPT-6.1 Sol, presenting it as “near-Astra intelligence for a fifth of the price.” The company describes the model as its most cost-efficient option for its performance, though the comparisons come from OpenAI’s own evaluations.
GPT-6.1 Sol costs $2 per million input tokens, $10 per million output tokens and $0.10 per million cached input tokens. GPT-6 Astra, by comparison, costs $10 per million input tokens, $50 per million output tokens and $1 per million cached input tokens. OpenAI also lists GPT-6 Luna at $0.10 per million input tokens, $0.50 per million output tokens and $0.01 per million cached input tokens.
OpenAI claims that GPT-6.1 Sol improves on GPT-6 Sol in coding, computer use and complex professional work. The company states that the model approaches GPT-6 Astra on several benchmarks at substantially lower cost.
The pricing advantage is particularly pronounced for cached inputs. OpenAI states that GPT-6.1 Sol’s $0.10-per-million-token rate is 95% below its standard input price and 50% below GPT-6 Sol’s cached input pricing.
OpenAI’s benchmark comparisons
OpenAI’s accompanying DeepSWE chart places GPT-6.1 Sol at scores ranging from roughly 64% to 74%, with plotted costs below $2 per task. GPT-6 Sol scores from approximately 37% to 69%, with costs below $3 per task, while GPT-6 Astra reaches roughly 67% to 75% at costs ranging from less than $2 to about $7.50 per task.
A separate GDP.pdf comparison places GPT-6.1 Sol between roughly 27% and 32%, with costs of about $0.30 to $0.45 per task. GPT-6 Sol scores from approximately 22% to 28% at similar costs. GPT-6 Astra reaches around 30% to 32%, but its plotted costs range from about $1.70 to $2.10 per task. The chart also includes Opus 5.5 with fallbacks, which scores between roughly 25.5% and 28.8% at costs ranging from about $0.75 to $1.55 per task.
On the OSWorld 2.0 offline set, OpenAI places GPT-6.1 Sol between approximately 59% and 71.5%, with costs below $1.50 per task. GPT-6 Sol ranges from about 44% to 64.5% at costs of roughly $1 to $3.50 per task. GPT-6 Astra scores between around 62% and 73.5%, but its listed costs range from approximately $2.70 to $9.50 per task.
The charts suggest that GPT-6.1 Sol occupies a position close to GPT-6 Astra on several tests while requiring considerably less spending per task. They remain company-produced comparisons rather than independent verification of the model’s performance.
Benchmarks
Paweł Huryn’s GPT-6.1 Sol benchmark fixed 44 of 105 planted bugs across two repositories at maximum effort, according to results published on September 29. The run carried a listed cost of $6.56 and took 112.4 minutes, while GPT-6 Astra Max fixed 45 bugs for $33.03 in 89.8 minutes.
GPT-6.1 Sol’s reported results
The Bug Hunt Bench evaluates coding models against planted defects in two repositories. Scores count bugs fixed and verified against a withheld answer key; unrelated defects are tracked separately and do not contribute to the score.
| Model and effort | Bugs fixed | Listed cost | Wall-clock time |
|---|---|---|---|
| GPT-6 Astra Max | 45 | $33.03 | 89.8 minutes |
| GPT-6.1 Sol Max | 44 | $6.56 | 112.4 minutes |
| GPT-5.6 Sol Max | 43.5 | $95.35 | 253.1 minutes |
| Opus 5.5 Max | 41.7 | $58.53 | 66.9 minutes |
| GPT-6 Sol Max | 29.3 | $9.33 | 63.4 minutes |
A high-turn, lower-cost run
The benchmark uses Antigravity CLI and counts model steps as turns, including tool calls, shell commands, file reads, edits, searches and the final response. Huryn reported that GPT-6.1 Sol Max averaged more than 1,400 turns across the two repositories—approximately seven times the number reported for GPT-6 Astra.
That extended process did not translate into the highest score, but it accompanied a substantially lower listed cost than the other Max runs in the same comparison. Huryn characterized GPT-6.1 Sol as a major improvement over GPT-6 Sol, which scored 29.3 bugs at maximum effort.
Alignment claims
OpenAI also reports “major improvements” over GPT-6 Sol in its alignment evaluations. The company claims that GPT-6.1 Sol is more transparent about its limitations and more reliable at following user intent and safety constraints, bringing it closer to GPT-6 Astra on those evaluations.
GPT-6.1 Sol is available immediately to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex.
Sources: OpenAI on X, Paweł Huryn on X
