Back

DeepSeek

DeepSeek V3.1 Terminus

ReasoningOpen weightsHugging Face
Intelligence
13.9
Rank 327 of 744 measured models
Coding
43.5
Rank 146 of 298 measured models
Output speed
40.8 t/s
Rank 362 of 403 measured models
Price/1M
$0.453
Rank 267 of 659 measured models
Context window
128K
Rank 773 of 1199 measured models

DeepSeek V3.1 Terminus (Non-reasoning) scores 13.9 on the Artificial Analysis Intelligence Index, ranking 327th of 744 measured models. The closest models on this index are Grok Code Fast 1 (14.1), DeepSeek V3.2 Exp (13.9), and Apriel-v1.5-15B-Thinker (13.8). At $0.453 per million tokens (3:1 input/output blended price), it is more expensive than DeepSeek V3.2 Exp ($0.315).

Artificial Analysis Indices
Non-reasoning
Intelligence13.9
Rank 327 of 744 measured models
Coding43.5
Rank 146 of 298 measured models
Math53.7
Rank 134 of 270 measured models
Agentic6.8
Rank 127 of 189 measured models
Reasoning levels
Each point is a reasoning level of this model: more reasoning often raises the score but costs more. Select a point to open that level.
Intelligence Indexmost efficient ↗Price / 1M tokens
Standard Benchmarks
MMLU Pro83.6%
Rank 52 of 349 measured models
GPQA Diamond75.1%
Rank 269 of 619 measured models
HLE8.7%
Rank 358 of 664 measured models
LiveCodeBench52.9%
Rank 139 of 343 measured models
AIME 202553.7%
Rank 134 of 270 measured models
IFBench41.2%
Rank 265 of 450 measured models
AA-LCR45.3%
Rank 345 of 575 measured models
TerminalBench Hard31.8%
Rank 105 of 432 measured models
Additional benchmarks7
Design Arena · 3d56.0%
Design Arena · Codecategories56.0%
Design Arena · Dataviz53.4%
Design Arena · Gamedev52.5%
Design Arena · Svg50.1%
Design Arena · Uicomponent59.5%
Design Arena · Website56.4%
Models at a similar level
The closest models on the Intelligence Index, one configuration per family.

Preferences and analytics

Rybbit: visits, performance, clicks, copies, forms and session replay (masked inputs), with your permission. Refusing does not limit the site.

Privacy

Allow navigation analytics?

With your permission, Rybbit measures visits, performance and interactions (clicks, copies, forms), including session replay, to improve BenchSift. Input values are masked in replays. Refusing does not limit the site.

Your choice is stored for 180 days. Change it at any time using “Privacy preferences” in the footer. Withdrawal reloads the page.

No choice stored: Rybbit is disabled.

Privacy