DeepSeek, released Apr 24, 2026

DeepSeek V4 Flashprice, context, benchmarks and release details

Provisional: not enough results to rank yet Open weights 53% confidence, 3 of 7 expected sources in

56.2

SI Score

#79 of 142 ranked models

Input, per 1M tokens: $0.30
Output, per 1M tokens: $1.20
Context window: 1M
Max output: 384K
Released: Apr 24, 2026

How the score breaks down

Coding 52.4

Math 74.5

Preference 76.1

Reasoning 69.5

Weights: reasoning 30%, math 15%, coding 40%, preference 15%. Results use fixed 0–100 scales before averaging, and thin evidence is pulled toward 50. Method si-v3-retained-evidence-2, computed Oct 9, 2026, 06:15 UTC.

Around it on the leaderboard

77 LongCat-2.0 56.6
78 MiMo-V2.5 56.4
79 DeepSeek V4 Flash 56.2
80 Fugu 55.9
81 MiMo-V2.6-Flash 55.8

Full leaderboard

Benchmark results

24 benchmarks, 24 results

Each row shows the best published result. Where a model was tested at several settings, such as reasoning effort, open the row to see each one. Hover or tap a value for its source.

GPQA Diamondreasoning 88.1% 88.1

Humanity's Last Examreasoning 34.8% 34.8

Humanity's Last Exam (with tools)reasoning 45.1% 45.1

LiveBench Reasoning: connectionsreasoning 89.5% 89.5

LiveBench Reasoning: consecutive eventsreasoning 59.9% 59.9

LiveBench Reasoning: logic with navigationreasoning 64% 64.0

LiveBench Reasoning: spatialreasoning 96% 96.0

LiveBench Reasoning: theory of mindreasoning 73.1% 73.1

LiveBench Reasoning: zebra puzzlesreasoning 49.3% 49.3

MMLU-Proreasoning 86.2% 86.2

LiveBench Math: AMPS Hardmath 98% 98.0

LiveBench Math: competition mathmath 97.1% 97.1

LiveBench Math: integralsmath 37% 37.0

LiveBench Math: olympiadmath 86.5% 86.5

LiveBench Math: simplifymath 54.0% 54.0

LiveBench Coding: code completioncoding 65.2% 65.2

LiveBench Coding: code generationcoding 73.2% 73.2

LiveBench Coding: JavaScriptcoding 54.5% 54.5

LiveBench Coding: Pythoncoding 35% 35.0

LiveBench Coding: TypeScriptcoding 23.3% 23.3

SWE-bench Procoding 52.6% 52.6

SWE-bench Verifiedcoding 79% 79.0

Terminal-Bench 2.0coding 56.9% 56.9

LMArena Textpreference 1432.1 elo 76.1

Normalization uses a fixed 0–100 scale for each unit, independent of other models. Compare evaluation conditions before reading a small gap as decisive. “Lab-reported” marks the provider's own published figure.

Details and sources

Open weights: Yes
License: not yet reported
Input modalities: text
First seen by SuperIndex: Oct 8, 2026
Coverage: 43% of expected source weight

Reported (3)

Official model cards via models.devOct 8, 2026
LiveBenchOct 8, 2026
LMArena / ArenaOct 8, 2026

Awaiting (4)

ARC Prize13% of weight
Epoch AI Benchmarking25% of weight
Humanity’s Last Exam13% of weight
Terminal-Bench6% of weight

Confidence rises as pending sources publish. Some sources never cover some models, so confidence reaches 100% at 80% of expected weight.