DeepSeek V4-Flash official release, and the frontier-model price war

Date: 31 Jul 2026 · OpenClaw 2026.7.1-2

DeepSeek pushed the official V4-Flash (build 0731) today, graduating it from preview. Same architecture and size as V4-Flash-Preview, just re-post-trained, but the reported agent/coding numbers are strong: Terminal Bench 2.1 at 82.7, DeepSWE 54.4, NL2Repo 54.2, Cybergym 76.7, and it now natively speaks the Responses API and is tuned for Codex. Notably DeepSeek claims this Flash release beats their own V4-Pro-Preview on benchmarks. This is a real frontier-model price war. A model posting these agentic results at Flash-tier pricing ($0.112 in / $0.224 out per 1M on DigitalOcean's current listing) squeezes everyone above it. Capability is climbing while price-per-token keeps falling.

We're running DeepSeek through DigitalOcean's Inference Engine, so we're waiting on DO to roll the 0731 weights. As of today their release notes show nothing about the refresh (latest DeepSeek-relevant entries predate it).