Back

Meta

Llama 4 Maverick

Open weightsHugging Face
Intelligence
9.3
Rank 448 of 744 measured models
Coding
16.3
Rank 240 of 298 measured models
Output speed
56.8 t/s
Rank 294 of 403 measured models
Price/1M
$0.405
Rank 243 of 659 measured models
Context window
1M
Rank 84 of 1199 measured models
Cost per task
$0.0382
Rank 30 of 193 measured models

Llama 4 Maverick scores 9.3 on the Artificial Analysis Intelligence Index, ranking 448th of 744 measured models. The closest models on this index are Gemini 2.0 Flash Thinking Experimental (Jan '25) (9.4), Gemini 2.5 Flash-Lite Preview (Sep '25) (9.3), and Mistral Large 3 (9.3). At $0.405 per million tokens (3:1 input/output blended price), it is cheaper than Mistral Large 3 ($0.75), more expensive than Gemini 2.5 Flash-Lite Preview (Sep '25) ($0.175). With an output speed of 56.8 t/s, it is slower than Mistral Large 3 (76.1 t/s).

Artificial Analysis Indices
Intelligence9.3
Rank 448 of 744 measured models
Coding16.3
Rank 240 of 298 measured models
Math19.3
Rank 212 of 270 measured models
Agentic0.6
Rank 183 of 189 measured models
Capability indexes
Performance in specific industries, measured by Artificial Analysis
Finance & Accounting9.0
Rank 193 of 218 measured models
Strategy & Ops7.6
Rank 180 of 218 measured models
Legal11.5
Rank 187 of 218 measured models
Healthcare & Medical9.1
Rank 80 of 82 measured models
Engineering11.3
Rank 199 of 217 measured models
Economics13.0
Rank 204 of 225 measured models
Standard Benchmarks
MMLU Pro80.9%
Rank 96 of 349 measured models
GPQA Diamond67.1%
Rank 350 of 619 measured models
HLE4.9%
Rank 488 of 664 measured models
LiveCodeBench39.7%
Rank 185 of 343 measured models
SciCode31.7%
Rank 207 of 231 measured models
MATH-50088.9%
Rank 78 of 202 measured models
AIME 202439.0%
Rank 72 of 194 measured models
AIME 202519.3%
Rank 212 of 270 measured models
Additional benchmarks6
Design Arena · 3d40.2%
Design Arena · Codecategories35.8%
Design Arena · Dataviz38.4%
Design Arena · Gamedev33.7%
Design Arena · Uicomponent40.8%
Design Arena · Website34.4%
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