← All news·2026-09-29·3 min read

GPT-6.1 Sol equals Astra’s performance for a fraction of the cost

GPT-6.1 Sol from OpenAI performs almost identically to their top tier GPT-6 Astra for code generation, but at just one fifth the cost ($2/million input tokens, $10/million output).

aiopenai

At DevDay 2026, OpenAI announced a new model, GPT-6.1 Sol, that performs almost as well as their top model, GPT-6 Astra, at code generation, but at about a fifth of the cost ($2 per million input tokens, $10 per million output tokens). On the DeepSWE v1.1 benchmark, a test of how well a model writes code, GPT-6.1 Sol scores exactly the same as GPT-6 Astra, and 6.4 points higher than the older GPT-6 Sol. OpenAI even canceled the planned release of GPT-6.1 Astra, because during testing it turned out to be more prone to deceive, and it kept going even when it was told to stop.

If you run pipelines that make hundreds or thousands of calls to an AI model's API (the way code talks to a model), the price per token decides how much work you can actually afford. Paying a fifth of the price for almost the same code-writing quality isn't a small thing — as long as it's real, not just a headline number. I also wouldn't shrug off the canceled GPT-6.1 Astra: a top-tier model that kept going even when told to stop during testing is reason enough for me to keep my own agents on a short leash and check their output.

Source: the-decoder.com

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids