InferenceBench's Nebius Token Factory dashboard surfaces recent 24-hour performance deltas for DeepSeek V4 Flash served on Nebius, including a throughput ↓58% reading on one window and a reliability recovery 96.8% → 100% on another. The same dashboard shows positive movement on related models routed via Nebius, includi Comparison readings also place DeepSeek V4 Flash on other providers in the same window: Scaleway throughput ↑359% with TTFT ↓39%, Fireworks AI throughput ↓16% with TTFT ↑17%, and Cortecs throughput ↑53%, alongside an OpenRouter reliability recovery 96.8% → 100%. For developers routing DeepSeek V4 Flash on Nebius Token

Artificial Analysis published a dedicated evaluation of DeepSeek V4 Flash 0731 on 2026-07-31, reporting an Intelligence Index score of 50 — a 10-point jump over the April 2026 DeepSeek V4 Flash (40) and one point behind GPT-5.6 Luna (max, 51). The model retains the same architecture and pricing as its predecessor, with The biggest gains are agentic: GDPval-AA v2 Elo rises from 1189 to 1559 (second-highest open-weights score behind Kimi K3's 1687 and ahead of GLM-5.2's 1510), Terminal-Bench 2.1 climbs 17 points to 79%, and τ³-Bench Banking improves 8 points to 31%. AA-Omniscience improves from -23 to -16 purely through reduced halluci

OpenRouter's page for deepseek/deepseek-v4-flash-0731 confirms the July 31, 2026 release of DeepSeek V4 Flash 0731 as a sparse mixture-of-experts model with 284B total parameters and 13B active, a 1M token context window, and a stated base price of $0.05 input / $0.16 output per 1M tokens with text in/out modalities. D As a router, OpenRouter enumerates many downstream inference hosts (OpenInference, Relace, DigitalOcean, DeepInfra, Together, Baseten, CoreWeave, Fireworks, Cloudflare, DeepSeek first-party, and others) with per-host latency, throughput, uptime, and pricing tiers for serving DeepSeek V4 Flash 0731. These figures reflec

DeepSeek's official API changelog documents two V4-family updates that frame the environment around V4 Flash 0731: on 2026-08-13 the GA release of DeepSeek-V4-Pro rolled out with materially enhanced agent capabilities (Terminal Bench 2.1 87.9, NL2Repo 61.5, Cybergym 83.3, Toolathlon-Verified 74.1, DSBench-FullStack 71. On 2026-08-21, DeepSeek shipped DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal vision understanding variant accessed via model='deepseek-v4-flash-vision-exp'. Its reported scores include Terminal Bench 2.1 83.9, NL2Repo 57.7, DeepSWE 59.3, DSBench-Hard 63.6, ApexBench Pass@1 36.5, Agents' Last Exam 27.3, Char