DeepSeek V4 Flash

DeepSeek V4 Flash sits within the DeepSeek Flash family as an open-weight language model positioned for coding and agentic development tasks. Umans AI publicly lists it among a curated set of frontier open coding models on its inference platform, alongside Kimi K3 and GLM 5.2, framing the offering around auditable open weights and pay-per-token access for real engineering workflows. The platform is built around the idea that recent open coding models have reached a quality bar where teams can ship production agentic code with them, and DeepSeek V4 Flash serves as one of the flagship backends for that promise.

On the ecosystem side, the model is catalogued under the NVIDIA NIM namespace at orgs/nim/teams/deepseek-ai/models/deepseek-v4-flash, indicating official NVIDIA packaging for accelerated inference. Community activity around an experimental V4 Flash Vision variant is visible on the NVIDIA Developer Forums, suggesting the model line is actively being extended beyond its core text capabilities. Umans AI integrates the model directly into popular coding agents such as OpenCode, Claude Code, Zed, and Cursor, letting developers reach it from the editors they already use while the provider handles the underlying GPU operations and reliability targets.

Umans AI Coding Planumans-deepseek-v4-flash-0731deepseek-flash

Open provider page Provider docs

Quick Info

Provider: Umans AI Coding Plan
Model key: umans-deepseek-v4-flash-0731
Release date: Jul 31, 2026
Last updated: Jul 31, 2026
Knowledge cutoff: 2025-05
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 393,215 tokens
Context window: 1,048,576 tokens

Latest news about DeepSeek V4 Flash

Umans AI Coding Plan

CoverageAnalysis

According to Artificial Analysis (July 31, 2026), DeepSeek V4 Flash 0731 scores 50 on the Artificial Analysis Intelligence Index, a 10-point jump over the April 2026 DeepSeek V4 Flash and 6 points above DeepSeek V4 Pro. The model retains the same architecture and pricing as its predecessor: a Mixture-of-Experts design The 0731 revision shows notable gains in agentic and long-context evaluations. DeepSeek V4 Flash 0731 reaches an Elo rating of 1559 on GDPval-AA v2 (up from 1189), climbs 17 points to 79% on Terminal-Bench 2.1, and gains 8 points to 31% on τ³-Bench Banking. Its AA-Omniscience Index improves by +7 to -16, driven entirel

Umans AI Coding Plan

CoverageBenchmark

DeepSeek shipped the official release of DeepSeek-V4-Flash-0731 on 31 July 2026, superseding the April 2026 preview at the same architecture and initial price point, according to a third-party recap citing DeepSeek's own model card. The 0731 build keeps the 284-billion-total / 13-billion-active Mixture-of-Experts confi The 0731 release delivers a substantial capability jump over the April preview: DeepSeek's model card puts it at 82.7 on Terminal-Bench 2.1 (up from 61.8), and Artificial Analysis moved the model from 40 to 50 on its Intelligence Index, placing it third in the open-weight field. The same recap notes DeepSeek later move