Meta, released Apr 5, 2025

Llama 4 Maverick 17B Instructprice, context, benchmarks and release details

Provisional: not enough results to rank yet Open weights 78% confidence 78 percent, Medium confidence, 6 of 9 expected sources in
30.0
SI Score
#141 of 142 ranked models
Input, per 1M tokens
not yet reported
Output, per 1M tokens
not yet reported
Context window
1Mmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Max output
16.4Kmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Released
Apr 5, 2025models.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗

How the score breaks down

Coding (weight 40 percent) 10.4
Math (weight 15 percent) 55.5
Preference (weight 15 percent) —
Reasoning (weight 30 percent) 9.2

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, 05:13 UTC.

Around it on the leaderboard

  1. 138 Llama-3.2-1B 32.1
  2. 139 GPT-4.1 nano 32.1
  3. 140 GPT-4o 31.2
  4. 141 Llama 4 Maverick 17B Instruct 30.0
  5. 142 GPT-4o mini 22.9

Full leaderboard

Benchmark results

10 benchmarks, 11 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.

ARC-AGI-1 (public eval)reasoning 7.1%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] Llama-4-Maverick-17B-128E-Instruct-FP8-togetherPublished Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 7.1
ARC-AGI-1 (semi-private)reasoning 4.4%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] Llama-4-Maverick-17B-128E-Instruct-FP8-togetherPublished Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 4.4
ARC-AGI-2 (public eval)reasoning 0%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] Llama-4-Maverick-17B-128E-Instruct-FP8-togetherPublished Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 0.0
ARC-AGI-2 (semi-private)reasoning 0%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] Llama-4-Maverick-17B-128E-Instruct-FP8-togetherPublished Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 0.0
GPQA Diamondreasoning 67.0%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Apr 8, 2025 Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 67.0
Humanity's Last Exam (Scale AI)reasoning 5.7%Humanity’s Last ExamPotential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. [variant] Published Apr 10, 2025 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 5.7
MATH Level 5math 73.0%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Apr 8, 2025 Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 73.0
OTIS Mock AIME 2024–2025math 20.6%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Apr 8, 2025 Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 20.6
Aider Polyglotcoding 15.6%Aider polyglotPublished source fact [variant] Aider polyglot; 225 cases; 2 attemptsPublished Apr 6, 2025 Retrieved Oct 9, 2026 · Apache-2.0
Open source ↗ 15.6
SWE-bench Pro (public)coding 5.2%Official model cards via models.devLab-reported; metric resolve rate; transcribed by MIT models.dev catalog; not independently evaluated [variant] public Retrieved Oct 9, 2026 · factual citation; MIT transcription
Open source ↗ 5.2 2 settings
  • Setting 1 5.2%Official model cards via models.devLab-reported; metric resolve rate; transcribed by MIT models.dev catalog; not independently evaluated [variant] public Retrieved Oct 9, 2026 · factual citation; MIT transcription
    Open source ↗
  • Setting 2 5.2%SWE-bench Pro (public)Published steward score [variant] Published Jan 27, 2026 Retrieved Oct 9, 2026 · factual citation
    Open source ↗
About SWE-bench Pro (public)

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
Yesmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
License
not yet reported
Input modalities
text, imagemodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
First seen by SuperIndex
Oct 8, 2026
Coverage
63% of expected source weight

Reported (6)

  • Aider polyglotOct 8, 2026
  • ARC PrizeOct 8, 2026
  • Epoch AI BenchmarkingOct 8, 2026
  • Humanity’s Last ExamOct 8, 2026
  • Official model cards via models.devOct 8, 2026
  • SWE-bench Pro (public)Oct 8, 2026

Awaiting (3)

  • LiveBench11% of weight
  • LMArena / Arena21% of weight
  • Terminal-Bench5% of weight

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

What changed

All releases · RSS feed

Alerts on this device

What to be alerted about
RSS feed