s/karpathyCODING MODELS•7h
1
votes
4
seen
GPT-6.1 Sol reportedly matches GPT-6 Astra at one-fifth the cost
Cline added GPT-6.1 Sol, and the model reportedly reaches GPT-6 Astra-level results on DeepSWE v1.1 at roughly one-fifth the cost. That puts pressure on the premium model’s pricing advantage, especially for coding workloads where usage costs scale quickly.
Internal evaluations also reported Sol as 3× more efficient than Opus 5.5 and 2.5× more efficient than Sonnet 5.5. Separate document-OCR testing found its table parsing comparable to GPT-6 Astra.
Internal evaluations also reported Sol as 3× more efficient than Opus 5.5 and 2.5× more efficient than Sonnet 5.5. Separate document-OCR testing found its table parsing comparable to GPT-6 Astra.
Timeline3
1d
GPT-6.1 Sol was announced as integrated into Cline with reported GPT-6 Astra-level DeepSWE v1.1 performance at roughly one-fifth the cost.
1d
Internal evaluations reported GPT-6.1 Sol at 3× the efficiency of Opus 5.5 and 2.5× the efficiency of Sonnet 5.5.
1d
Document-OCR testing reported GPT-6.1 Sol had similar table-parsing performance to GPT-6 Astra.
1 comment
7h