September 29, 2026
GPT-6.1 Sol tasks cost 4 times less: a Terminal-Bench measurement against Astra
On Sep 29, GPT-6.1 Sol became available in Codex and ChatGPT Work on paid plans, but not in regular ChatGPT. In the API, one million input tokens cost $2, cached tokens cost $0.10, and output tokens cost $10. The model has a 1.05 million-token context window, and responses can be up to 128 thousand tokens long.
Dan Shipper
@danshipper
OpenAI launches GPT-6.1 Sol: Astra-level performance at one-fifth of the price!!
· 5.3K views
The model is called with the ID `gpt-6.1-sol`; OpenAI recommends the Responses API for its tools. After 272 thousand input tokens, input and cache pricing doubles, while output pricing rises 1.5x for the entire request. There is no free API tier: Tier 1 provides 500 RPM and 500 thousand TPM, while Tier 5 provides 15 thousand RPM and 40 million TPM.
A new tier. GPT-6 Sol was previously part of the lineup. OpenAI now says GPT-6.1 Sol matched GPT-6 Astra on DeepSWE v1.1 and surpassed GPT-6 Sol's best result by 6.4 percentage points with lower effort and cost. On Terminal-Bench Science 0.1, the average task at maximum reasoning cost $5.47, versus $23.80 for Astra and $23.21 for Opus 5.5; Astra posted the best result there at 68.1%.
GPT-6.1 Sol supports the following reasoning modes: low, medium, high, xhigh, and max.
