Cognition announced the release of SWE-1.7, a new AI model trained from Kimi K2.7 and available at 1,000 tokens per second. The company claims its performance on benchmarks approaches that of GPT-5.5 and Opus 4.8, but at a significantly lower cost.
The model is part of a broader push by Cognition to deliver high-performance AI at reduced operational expenses. The company did not disclose specific benchmark scores or cost comparisons, making independent verification difficult.
SWE-1.7 operates at a rate of 1,000 tokens per second, a speed that could enable real-time applications in coding, analysis, and other compute-intensive domains. No pricing details have been released yet.
If the performance claims hold, this could pressure larger AI providers to lower their own prices or improve efficiency. The model may also accelerate adoption of AI in cost-sensitive enterprise environments.
However, without third-party validation, the claims remain unsubstantiated. Analysts will likely await independent benchmarks before drawing firm conclusions.