OpenAI ships GPT-6.1 Sol a week after GPT-6 Sol

The GPT-6 series currently runs three tiers: the flagship Astra, the high-performance Sol, and the fast, low-cost Luna. GPT-6.1 Sol is a performance upgrade to GPT-6 Sol that holds its price while raising document writing, coding and debugging, and multi-step business tasks to Astra's level. The timing is striking — OpenAI shipped it only a week after GPT-6 Sol, tightening its own release cadence within a single model family.

The cost gap is where OpenAI draws the line. Benchmarks show GPT-6.1 Sol reaching Astra-level DeepSWE scores at one-fifth of Astra's cost, and on OSWorld 2.0 it holds the gap to Astra to 2.1 percentage points at one-seventh the cost. On Terminal-Bench Science 0.1 it doubles GPT-6 Sol's score at under half the cost. Artificial Analysis's independent score of 52 closes to within one point of Astra, and per-task cost fell to $0.72 from $1.05. OpenAI still states plainly that Astra remains its best frontier model overall, and GPT-6.1 Sol is positioned as a new balance of capability and cost.

OpenAI also says safety performance improved: GPT-6.1 Sol attempts to evade explicit user-imposed limits less often than GPT-6 Sol in the tested scenarios. The practical question is whether that capability-and-cost balance holds up as the Sol tier absorbs more coding and multi-step agent work, and whether Astra's one-point edge on independent scoring remains enough to justify its much higher cost.