GPT-6.1 Sol: near-Astra performance at Sonnet 5.5’s price
OpenAI’s upgraded mid-tier model. What it scores, what a task costs, where you can use it, and when to pick it over GPT-6 Astra or Claude Sonnet 5.5.
The short answer
A strong, cheap default for coding agents and document work—OpenAI’s own numbers put it close to Astra at a fraction of the cost. Test it against Sonnet 5.5, which lists at the same price.
OpenAI’s announcement (opens in a new tab)1 · Performance · vendor-reported
What OpenAI says it can do
- Coding (DeepSWE v1.1)
- Comparable to GPT-6 Astra at about one-fifth of the cost per task; 6.4 points above GPT-6 Sol’s best score.Vendor-reported
- Documents (GDP.pdf)
- Scores above Claude Opus 5.5 (with fallback) at every reasoning setting tested, at under half the cost per task.Vendor-reported
- Business workflows (AutomationBench)
- 2.2 points above Opus 5.5 at medium effort, at about one-third of the cost.Vendor-reported
- Computer use (OSWorld 2.0)
- Within 2.1 points of Astra at maximum effort, at about one-seventh of the cost per task.Vendor-reported
- Science (Terminal-Bench Science 0.1)
- Average cost per task $5.47 vs $23.21 for Opus 5.5 and $23.80 for Astra; Astra still leads at 68.1%.Vendor-reported
- Factual errors (hard prompts)
- 4.1% at extra-high effort vs 4.0% for Astra and 4.5% for GPT-6 Sol.Vendor-reported
All figures are OpenAI’s own evaluations; competitor results are taken from public reports, and settings differ. Independent results will follow—we will add them here.
2 · Price and task cost
What it costs
- API input
- $2.00 per 1M tokens
- Cached input
- $0.10 per 1M tokens (95% lower)
- API output
- $10.00 per 1M tokens
- Faster option
- GPT-6.1 Sol Ultrafast in the API: near-Astra quality, up to 8× Astra’s speed, at roughly Astra’s old API cost
3 · Where you can use it
Availability
- ChatGPT
- Plus, Pro, Business, Enterprise and Edu—in ChatGPT Work and Codex, not in regular chat yet
- API
- gpt-6.1-sol
- Other apps
- Apps that accept “Sign in with ChatGPT” plan usage may route to it; check each app
4 · Our read · opinion
When to use it
Pick GPT-6.1 Sol
Coding agents, PDF-heavy professional work and multi-step business workflows where cost per task matters.
Pick GPT-6 Astra
The hardest science and open-ended problems—OpenAI itself says so—or when you need dots (they run on Astra).
Test against Claude Sonnet 5.5
Same $2 / $10 list price. Anthropic reports Sonnet 5.5 near Opus 5.5; OpenAI reports Sol above Opus 5.5 on documents. Run one real task on both.