GPT-6.1 Sol vs GPT-6 Astra: Which Fits Your Workload?
GPT-6.1 Sol lists at $2/$10 per million tokens, one-fifth of Astra's price, and OpenAI says it nearly matches Astra on coding and computer use.
OpenAI introduced GPT-6.1 Sol at DevDay 2026 on September 29, a week after launching GPT-6 Sol and Luna, and the notable part is the price-to-capability ratio rather than a new ceiling. The upgrade to GPT-6 Sol lists at $2 per million input tokens and $10 per million output, one-fifth of GPT-6 Astra's rates, while OpenAI reports that it comes close to Astra on agentic coding, computer use and professional work. For teams choosing between OpenAI's tiers, the practical question becomes where the remaining gap to Astra still matters.
- Pricing: $2 per million input tokens and $10 per million output, one-fifth of Astra's standard rates, plus $0.10 cached input, half of GPT-6 Sol's $0.20
- OpenAI-reported performance: matches Astra on DeepSWE v1.1 and lands within 2.1 points on OSWorld 2.0, at roughly one-fifth and one-seventh of Astra's cost
- Availability: ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu plans from launch day, and the API as
gpt-6.1-sol - Specs: a 1,050,000-token context window, 128,000 maximum output tokens and an April 30, 2026 knowledge cutoff
How Does GPT-6.1 Sol Compare to GPT-6 Astra?
These are OpenAI's own results, and the company says its competitor figures come from public reports:
| Evaluation | GPT-6.1 Sol result | Cost note |
|---|---|---|
| DeepSWE v1.1 (real-codebase engineering) | Matches Astra; 6.4 points above GPT-6 Sol's best score, at lower effort | About one-fifth of Astra's cost |
| OSWorld 2.0 offline set (computer use, max effort) | 7 points above GPT-6 Sol; 2.1 points below Astra | About one-seventh of Astra's cost per task |
| Terminal-Bench Science 0.1 (max effort) | More than double GPT-6 Sol's score; Astra leads at 68.1% | $5.47 per task versus $23.80 for Astra |
| AutomationBench 1.0.6 (medium effort) | 4.8 points above GPT-6 Sol and 2.2 above Opus 5.5 | About one-third of Opus 5.5's cost |
| Factual errors on hard prompts (low effort) | 7.7% of answers, down from 11.4% | Within 1.9 points of Astra across settings |
Astra keeps the lead where difficulty peaks: OpenAI says it still posts the highest Terminal-Bench Science score among the models it tested and recommends it for the hardest scientific research tasks. The factuality test uses deliberately hard prompts, drawn from conversations where users had flagged an earlier model's mistake, so OpenAI says it does not reflect typical use.
We found no independent benchmark results at the time of writing; Artificial Analysis, which scored GPT-6 Sol, had not yet published any. Treat these comparisons as OpenAI's claims until outside testing arrives.
What Does GPT-6.1 Sol Cost?
List prices are $2 per million input tokens and $10 per million output, the same as GPT-6 Sol. The change is cached input, which falls to $0.10 per million from $0.20, just 5% of the uncached rate and a saving for agents that resend the same context on every step. Astra lists at $10 input, $1 cached and $50 output. The $2 and $10 pairing also matches Claude Sonnet 5.5, which launched a day earlier and is covered in our Sonnet 5.5 versus Opus 5.5 comparison.
Three modifiers apply to those list prices. Prompts above 272,000 input tokens, inside the 1.05M-token window, are billed at twice the input and cache rates and 1.5 times the output rate for the whole request; Batch and Flex requests cost half of standard rates; and Fast mode costs double.
Where GPT-6.1 Sol Is Available Today
OpenAI says the model is live now for Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, though not yet in Chat, and in the API as gpt-6.1-sol. An Ultrafast option due in the coming days offers up to eight times faster token generation than standard speed in Codex.
What Developers Should Know Before Switching
- Reasoning effort supports low, medium (the default), high, xhigh and max; the none and minimal settings are not accepted
- Tool calling runs through the Responses API and Chat Completions works without tools, so a GPT-6 Sol setup that relied on none-effort function calling in Chat Completions will need to move to the Responses API
- Inputs can be text or images and output is text; US and EU data residency are both supported, though Fast mode is not available with EU residency
OpenAI's documentation suggests testing Sol against Astra on your own tasks to weigh quality against cost. Our guides to GPT-6 Sol, Luna and Astra and Astra's pricing and computer-use skills cover the rest of the family.
How OpenAI Describes the Safety Evaluations
OpenAI reports alignment gains over GPT-6 Sol that bring the model closer to Astra. It says GPT-6.1 Sol is more upfront about its limitations, fails less often at flagging a broken search tool, respecting explicit restrictions and avoiding unauthorized outcomes in agent tasks, and showed no attempts to bypass an automated safety reviewer. These tests deliberately probe hard situations rather than typical use, and the details are in OpenAI's system card addendum. More model launches land in our AI coverage.
Sources: OpenAI: Introducing GPT-6.1 Sol — September 29, 2026; OpenAI API docs: GPT-6.1 Sol — accessed September 29, 2026; TechCrunch — September 29, 2026; The Decoder — September 29, 2026; AFP via The Korea Times — September 30, 2026. Every performance figure above comes from OpenAI's own evaluations.
More AI Stories
Strands Decider 2B: What Amazon's Free Decision Model Does
Strands Decider 2B is a free 2B-parameter decision model that picks options in about 115ms on a single GPU. Here is how it works and where it fits.
Meta Muse Gadgets SDK: Build Your Own AI Agent Hardware
Meta's open-source Muse Gadgets SDKs bring its Muse agent to Raspberry Pi and ESP32 builds, and 5,000 free Home Link dongles are going to subscribers.
AstaBrief 8B: Ai2's Open Model for Cited Science Reports
AstaBrief 8B is Ai2's Apache 2.0 open model that writes cited research reports in 51 seconds, 3.5x faster than its Claude pipeline. Here is how it works.