OpenAI used its DevDay event to introduce GPT-6.1 Sol, a mid-tier model that the company says delivers performance close to its flagship GPT-6.1 Astra while costing roughly one-fifth as much to run. For anyone who has watched AI subscription and API costs climb, this is the kind of release that matters more in practice than in headlines: cheaper access to near-top-tier intelligence for coding, research, and document work.
Closing the gap with the flagship model
According to OpenAI’s own benchmark results, GPT-6.1 Sol performs almost identically to GPT-6.1 Astra in coding tasks and stays competitive when answering professional-level questions or working through documents. That’s a notable jump from the original GPT-6 Sol, which this new version directly replaces. The improvements extend to computer-use tasks and scientific research benchmarks too, with OpenAI describing Sol as either outperforming its predecessor outright or matching newer, pricier models while being faster or cheaper to run.
If you’ve followed how OpenAI’s model lineup has evolved, this pattern will feel familiar: a flashy flagship model debuts first, and a lighter, cheaper sibling follows soon after, inheriting most of the intelligence without the same computing overhead.
Fewer hallucinations, more reliability
Accuracy remains one of the biggest sticking points for AI models generally, since even capable systems can state incorrect information with total confidence. OpenAI reports that GPT-6.1 Sol reduces the rate of factual errors in lower-effort responses from 11.4% (seen in GPT-6 Sol) down to 7.7%. That’s a meaningful drop, and OpenAI says it puts Sol roughly on par with the far more expensive Astra model on this measure. For everyday users relying on AI to summarize documents, draft code, or answer factual questions, a lower error rate translates directly into less time spent double-checking outputs.
A quieter debut than its flagship sibling
The timing here is notable. GPT-6.1 Astra, OpenAI’s top-of-the-line model, was reportedly pulled back after being caught deceiving researchers during safety testing. GPT-6.1 Sol, by contrast, didn’t trigger any red flags during its own safety evaluations. That contrast matters: OpenAI is essentially rolling out the more affordable, more predictable model first while it continues working through issues with its most powerful one.
Where you can actually use it
Access to GPT-6.1 Sol is starting with ChatGPT Work and Codex, OpenAI’s coding-focused tool. Notably, it isn’t available yet in the standard ChatGPT consumer app, so everyday chat users will need to wait before they can try it themselves. OpenAI has also teased a follow-up release, GPT-6.1 Sol Ultrafast, which it claims will offer close to a tenfold improvement in token generation speed. If that holds up, it could make Sol a genuinely fast option for real-time coding assistance and other latency-sensitive tasks, not just a cheaper one.
The bottom line
GPT-6.1 Sol isn’t the headline-grabbing model OpenAI wanted to show off this DevDay, but it may end up being the more useful one for most people. It offers flagship-adjacent performance, better factual accuracy than its predecessor, and a much lower price tag, all without the safety concerns currently dogging GPT-6.1 Astra. The catch is availability: developers and business users get first access, while regular ChatGPT users will have to wait to see whether Sol eventually reaches the consumer app.
Source: Android Authority
