BenchLeader
Menu
LeaderboardsModelsProvidersBenchmarksCompareReleasesBlogMethodology
MetaOpen weights: meta-llama/Llama-4-Maverick-17B-128E-Instruct

Llama 4 Maverick

Best configuration ranks #456 of 610 on the BenchLeader Index at 42.6 ±3.8. Last measured 2 Sept 2026. Released 5 Apr 2025.

Blended price
$0.400/M
$0.250 in · $0.850 out
Output speed
91 tok/s
First answer
0.85 s
first token 0.58 s
Context
1.0M
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index43
  2. Reasoning39
  3. Coding32
  4. Agents & tools46
  5. Maths45
  6. Knowledge38
  7. Instruction following37
  8. Human preference49
  9. Multimodal47
  10. Long context44
  11. Composite41

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

BenchmarkdefaultSource
GPQA Diamond67.0%#162Epoch AI Benchmarking Hub
Humanity's Last Exam5.7%#36Scale AI / CAIS
SimpleBench27.7%#68SimpleBench
LMArena Hard Prompts1338#201LMArena
GPQA Diamond (AA)not in index67.1%#309Artificial Analysis
Humanity's Last Exam (AA)not in index4.9%#387Artificial Analysis
Kagi LLM Benchmark55.9%#61Kagi LLM Benchmark
ARC-AGI-14.4%#180ARC Prize
ARC-AGI-20.0%#174ARC Prize

Coding

BenchmarkdefaultSource
SciCode33.1%#138SciCode
WeirdML24.5%#132WeirdML
LMArena Coding1373#197LMArena
SciCode (AA)not in index31.7%#144Artificial Analysis
Aider Polyglot15.6%#39Aider polyglot leaderboard
SWE-bench Verified (bash only)21.0%#41SWE-bench
SWE-bench Verified (any scaffold)not in index21.0%#61SWE-bench

Agents & tools

BenchmarkdefaultSource
Terminal-Bench Hard6.8%#240Artificial Analysis
τ²-Bench Telecom (AA)not in index17.8%#334Artificial Analysis
BFCL Overall37.3%#38Berkeley Function Calling Leaderboard

Maths

BenchmarkdefaultSource
OTIS Mock AIME20.6%#203Epoch AI Benchmarking Hub
MATH Level 573.0%#36Epoch AI Benchmarking Hub

Knowledge

BenchmarkdefaultSource
AA-Omniscience-41.9#264Artificial Analysis
MultiNRC8.4%#42Scale AI SEAL

Instruction following

BenchmarkdefaultSource
IFBench43.0%#225Artificial Analysis
TutorBench40.2%#26Scale AI SEAL

Human preference

BenchmarkdefaultSource
LMArena Text1327#204LMArena

Multimodal

BenchmarkdefaultSource
LMArena Vision1142#99LMArena
MMMU-Pro62.1%#164Artificial Analysis
VISTA38.3%#41Scale AI SEAL
MMMU (validation)73.4%#18MMMU

Long context

BenchmarkdefaultSource
Fiction.LiveBench 120k36.4%#29Fiction.live
AA-LCR50.0%#253Artificial Analysis

Composite

BenchmarkdefaultSource
Epoch Capabilities Indexnot in index132.2#126Epoch AI Benchmarking Hub
AA Intelligence Index9.3#333Artificial Analysis

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

ProviderSpeedFirst tokenInput $/MOutput $/MContextQuantisation
Google Vertex89 tok/s0.65 s$0.350$1.15524k–
NovitaAI71 tok/s0.56 s$0.270$0.8501.0Mfp8
DeepInfra65 tok/s0.41 s$0.200$0.8001.0Mfp8
Parasail41 tok/s0.58 s$0.350$1.00524kfp8
DigitalOcean12 tok/s0.62 s$0.200$0.696128k–

Price history

Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.

$0.00$0.239$0.479$0.718$0.957Jun 26Jul 26Aug 26
input outputnow $0.200 in · $0.696 out

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0004–4.1 s
Summarise a 30-page report12,000 / 600$0.0035–7.4 s
Code edit6,000 / 1,500$0.0028–17.3 s
Agentic coding session60,000 / 4,000$0.018–44.6 s
Structured extraction2,000 / 200$0.0007–3.0 s

See also

Best under $2/MBest under $0.50/MBest open weightsCheapest 1M contextAll Meta models

Data as of 9 Sept 2026. Compare with another model.