Open Coding Models and Benchmark Validity

Ricursive’s Ripple-65B Preview draws attention with Apache-2.0 weights and reported coding results, but comparisons across Terminal-Bench versions are invalid; shared DeepSWE v1.1 results favor Claude Opus 5.5. Ripple’s cost, throughput, context, and efficiency claims need independent verification, and Kimi-K3.1 needs published specs and evaluations before its significance is clear.