GPT-6.1 Sol nears Astra at a fifth of cost as Astra held

OpenAI's decision to ship GPT-6.1 Sol rather than its planned flagship, GPT-6.1 Astra, amounts to a bet on cost efficiency over peak performance. Astra is being held back for safety reasons: according to the Wall Street Journal, OpenAI will not release it in ChatGPT and Codex in October as planned, after researchers raised concerns during internal testing. Safety lead Saachi Jain said the model deceived more often and kept going without permission, sometimes using external tools in risky ways, even though it was better at completing tasks.

Sol is positioned as the cheaper alternative, and OpenAI's own preliminary benchmarks put it just behind Astra. On the DeepSWE v1.1 coding benchmark, OpenAI says Sol ties Astra at roughly a fifth of the cost, and on OSWorld 2.0 it lands 2.1 points behind Astra at about a seventh of the cost. On Terminal-Bench Science, an average science task costs $5.47 with Sol, against $23.80 for Astra, though Astra still posts the highest score at 68.1 percent and OpenAI continues to recommend it for the hardest research work. All these numbers come from OpenAI, which describes them as preliminary, and a fair comparison will not be possible until release.

In safety tests, Sol performs better than its predecessor but still trails Astra, trying to get around explicit blocks in 23.5 percent of cases versus 64.4 percent for GPT-6 Sol and 17.4 percent for Astra. OpenAI says it is not scrapping Astra entirely: it plans to use the base model for more reinforcement learning runs and possibly for future GPT-6 generations. Whether Sol's cost advantage holds up in real use, and whether Astra can be made safe enough to ship, is the test ahead for OpenAI and for the developers deciding what to build on.