DeepSeek V4.1 Flash test: official API hits 99% cache vs 92-96% on OpenCode Go
teortaxesTex · x · 2026-10-23
A hands-on comparison of DeepSeek V4.1 Flash 1017 via the official API vs the OpenCode Go reseller channel: the official API was consistently faster (the gap widens on longer tasks), cache hit rates were 99%+ vs 92–96%, and OpenCode produced more timeouts and wasted runs, negating its advertised 8x inference edge. Commenter teortaxes notes this suggests OpenCode's Whale tokens are a separate lower-cost agreement rather than subsidized consumer API, which may make the model sustainable.
More from Infra
- Qdrant benchmarks Cohere Embed 5 Pro: one dense vector per PDF page vs ColPali — qdrant_engine · 2026-10-23
- DeepInfer's math shows trailing SOTA maximizes profit under compute constraints — examachine · 2026-10-23
- Benchmark: Direct gcloud CLI beats Skills and MCP for token-cheap agent deploys — rseroter · 2026-10-23
- Google funds one of the largest virtual power plants to free up electricity for data centers — Grady_Booch · 2026-10-23
- IQ2_XXS quant + 64k context makes a 27B model spin forever instead of coding — mailto_devnull · 2026-10-23
- Chris Lattner announces new alliance to push open, community-driven AI compute ecosystem — clattner_llvm · 2026-10-23