modelOct 11, 2026Meridian AI
Meridian-2 tops SWE-bench Verified with a 71% agentic solve rate
The new flagship pairs a 1M-token context window with a trained agentic loop. Independent replications are still pending, but the eval harness is open.
Source link coming soon — this summary is from our morning brief.
More headlinesSummarized by the Neuricon wire. Every claim links back to its primary source where available.
Comments
Loading comments…