Near-Frontier AI Just Got Five Times Cheaper
OpenAI's new GPT-6.1 Sol delivers almost the same results as its most powerful model at a fifth of the cost — and that changes the math for anyone automating work with AI.
- $2
- per million input tokens
- 1/5
- of GPT-6 Astra's price
- ~75%
- DeepSWE v1.1 benchmark score
- $0.30
- per automated task (AutomationBench)
The cost of running powerful AI keeps falling. On September 29, OpenAI launched GPT-6.1 Sol — a model that nearly matches its top model, GPT-6 Astra, on coding and automation tasks. The key difference: it costs one-fifth of GPT-6 Astra — $2 per million input tokens versus $10, and $10 per million output tokens versus $50.
To understand why this matters, think about the cost of automating real business tasks. A benchmark called AutomationBench — which measures how well an AI can complete a full task from start to finish — put GPT-6.1 Sol at about $0.30 per automated task. When tasks cost thirty cents instead of a dollar or more, the number of things worth automating grows fast.
The model also handles a very large amount of text at once. Its context window — the amount it can read in a single call — holds about one million tokens, or roughly 750,000 words. Entire contracts, codebases, or research files fit in one call. That makes it practical for the kind of complex, multi-step work that was too costly or slow before.
On software engineering tests, Sol scores around 75% on a standard benchmark called DeepSWE v1.1 — matching GPT-6 Astra and beating the previous Sol model by more than six percentage points. A separate accuracy test found it makes fewer errors than its predecessor at lower reasoning settings — down from 11.4% to 7.7%.
The pattern is steady: every few months, what cost $10 costs $2. What used to require the most expensive model now works at mid-tier prices. If your team ran the numbers on an AI automation project six months ago and passed, that calculation is worth redoing today.