GeekRadar / The comparison desk

COMPARE.

AI models

deepseek

DeepSeek V4 Flash 0731

AI model Released Jul 31, 2026

Context window: 1,000,000 tokens
Maximum output: 384,000 tokens
Open weights: Yes

Capability & preference

General capability · ECI

A cross-benchmark capability estimate. Index points, not a percentage; intervals show uncertainty.

DeepSeek V4 Flash 0731 154.9^* DeepSeek V4 Flash 0731 · 95% interval 152.6–156.8

Model performance

Radar task score: four tests, equal weight. An experimental summary of these tasks, not a universal intelligence rating.

DeepSeek V4 Flash 0731

Tested configuration Radar task score

54.1^* / 100

4/4 core tests · experimental v1

6 benchmark results for this configuration

Scientific reasoning

GPQA Diamond · 25%

Expert-written science questions; strict multiple-choice scoring. Chance accuracy is 25%.

DeepSeek V4 Flash 0731 91.0^*% Approx. 95% interval: 87.6–94.5% · 2026-08-02

Novel problem solving

Mystery Game Puzzles · 25%

Best-next-move puzzles in an undisclosed game variant. A narrow reasoning test, not general intelligence.

DeepSeek V4 Flash 0731 34.0^*% Approx. 95% interval: 24.7–43.3% · 2026-08-05

Advanced mathematics

FrontierMath Tiers 1-3 · 25%

Private v2 mathematics problems with Python tools. Earlier versions and Tier 4 are excluded.

DeepSeek V4 Flash 0731 57.5^*% Approx. 95% interval: 51.8–63.3% · 2026-08-02

Factual answers

SimpleQA Verified · 25%

Correct factual answers under a guessing prompt, graded by a model. Not a hallucination or abstention score.

DeepSeek V4 Flash 0731 33.6^*% Approx. 95% interval: 30.7–36.6% · 2026-08-27

Competition mathematics

OTIS Mock AIME 2024-2025 · Separate result

Competition problems with model-extracted answers checked against an exact key. Earlier grading procedures are excluded.

DeepSeek V4 Flash 0731 94.4^*% Approx. 95% interval: 90.4–98.5% · 2026-08-02

Research mathematics

FrontierMath Tier 4 v2 · Separate result

The hardest private FrontierMath problems, corrected v2 set, with Python tools. Separate from Tiers 1-3 in the core score.

DeepSeek V4 Flash 0731 24.4^*% Approx. 95% interval: 11.1–37.7% · 2026-08-02

Bars show reported task scores. Whiskers are approximate 95% intervals where standard errors are available. Different reasoning budgets are not equal-cost tests. Missing results are never zero.

Run it locally

Inference · memory estimate

DeepSeek V4 Flash 0731

Not enough verified architecture information for a reliable VRAM estimate.

API cost

USD / month

Input tokens / monthOutput tokens / month

Standard text-token estimate. Excludes taxes, caching, batch discounts, reasoning and multimodal surcharges. Provider limits and regional availability may differ.

DeepSeek V4 Flash 0731

Provider offer

DeepSeek V4 Flash 0731 / Pricing tiers 1

Provider / scope	Rate	Currency	Amount / unit
DeepInfradeepseek-ai/DeepSeek-V4-Flash-0731Standard	Cached input	USD	0.015 / 1M tokens
	Input	USD	0.06 / 1M tokens
	Output	USD	0.18 / 1M tokens

Capabilities

Context window: 1,000,000 tokens
Maximum output: 384,000 tokens
Reasoning: Yes
Tool calling: Yes
Open weights: Yes
Knowledge cutoff: 2025-05
Model license: MIT

Input modalities

text

Output modalities

text

Processing & platform

Architecture: deepseek_v4

* Provisional data