DeepSeek V4 Flash 0731 (Nebius)

DeepSeek V4 Flash 0731 (Nebius) is positioned as a lightweight member of the DeepSeek V4 family, distilled for speed while keeping the large context window that the V4 line is known for. It is offered as an open-weight text-to-text model, which means developers can self-host or fine-tune from the released weights while still relying on hosted inference through supported channels. The Nebius listing advertises a 1,024,000-token context window and matching 1,024,000-token maximum output, making it well suited to workloads that need to ingest and reason over very long documents such as code repositories, research corpora, or extended transcripts in a single pass.

The model is shaped for practical production use rather than purely academic benchmarks. Support for tool calling, structured output, and adjustable temperature lets it slot into agentic pipelines, retrieval-augmented systems, and JSON-generating backends, while its open-weight availability lowers the barrier to customization and on-premise deployment. Compared with larger V4 variants like the DeepSeek V4 Pro sibling, the Flash tier targets lower-latency, cost-sensitive tasks where throughput matters more than peak reasoning depth, making it a flexible general-purpose workhorse for everyday language and reasoning workloads.

Eden AInebius/deepseek-ai/DeepSeek-V4-Flash-0731deepseek-flash

Open provider page Provider docs

Quick Info

Provider: Eden AI
Model key: nebius/deepseek-ai/DeepSeek-V4-Flash-0731
Release date: Jul 31, 2026
Last updated: Jul 31, 2026
Knowledge cutoff: 2025-05
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.14
Output token cost: $0.28

Limits

Output tokens: 384,000 tokens
Context window: 1,024,000 tokens

Transparent token rates

Compare DeepSeek V4 Flash 0731 (Nebius) pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about DeepSeek V4 Flash 0731 (Nebius)

Eden AI

Coverage

The bestgeneralaiagents guide explicitly covers DeepSeek V4 Flash 0731, noting DeepSeek flipped V4-Flash from preview to official on July 31, 2026, and frames the 0731 checkpoint as delivering near-flashship agent performance at substantially lower cost. It confirms the 284B-total / 13B-active MoE design, 1M-token cont Official agent benchmark numbers from the July 31 release (tested with the DeepSeek Harness in minimal mode at max effort) are tabulated against V4-Pro GA: Terminal-Bench 2.1 at 82.7 vs 87.9, Cybergym 76.7 vs 83.3, Toolathlon (verified) 70.3 vs 74.1, DSBench-FullStack 68.7 vs 71.1, DeepSWE 54.4 vs 62.7, NL2Repo 54.2 vs

Eden AI

CoverageBenchmark

OpenLLMStack's spec page names DeepSeek-V4-Flash-0731 directly and confirms it as the official release superseding the April 2026 preview, entering public beta on the API on July 31, 2026 with unchanged architecture and parameter count — all gains coming from a redone post-training run. The architecture is the DeepSeek Two 0731-specific details are documented: DSpark speculative decoding ships attached to the released weights, targeting layers 40–42 and enabled with a single flag; and a selectable reasoning effort parameter accepts low, high, or max (published scores use the higher settings, with DeepSeek recommending up to 384K outp

Eden AI

CoverageRelease Notes

Vorp Labs' August 17, 2026 release review covers DeepSeek-V4-Flash-0731 explicitly by name and date (July 31, 2026), describing it as the official release of DeepSeek's fast MoE line superseding the April 24 preview through a re-post-training pass. The page confirms the same 284B-total / 13B-active MoE architecture (30 The review also flags migration traps relevant to developers: every peak/off-peak tier is an increase over the retired $0.14/$0.28 flat prices, making pre-repricing cost models stale; OpenRouter's dated 0731 route retains the retired flat rates while the rolling slug pins the April preview snapshot at $0.0798 / $0.1596