Llama 4 Maverick vs Llama 4 Scout
Wins 5 of 7 areas
Coding · Reasoning · Images and charts · Long documents · Following instructions
Wins 0 of 7 areas
—
Llama 4 Maverick is the stronger all-rounder.Llama 4 Scout is cheaper.
Scores updated · 30 tests both models report · How we compare
Where each one wins
Tests won in each of the seven areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software40Llama 4 Maverick4 of 4 tests
- Images and chartsUnderstanding pictures, charts and video30Llama 4 Maverick3 of 3 tests
- ReasoningHard problems that need careful thinking20Llama 4 Maverick2 of 4 tests · 2 ties
- Long documentsFinding answers in very long texts10Llama 4 Maverick1 of 1 test
- Following instructionsDoing exactly what it is asked10Llama 4 Maverick1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 3 ties
- FactsGetting facts right instead of making them up11Even1 each
Long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Llama 4 Scout costs 22% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Llama 4 Maverick pulls ahead
- Reasons across sets of long documentsAA-LCR+22.3points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+11.9points ahead
- Code for real scientific research problemsSciCode+10.4points ahead
Where Llama 4 Scout pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+9.5points ahead
Every test, side by side
All 30 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingLlama 4 Maverick
- SWE-bench VerifiedLlama 4 Maverick by 11.9219.1+11.9
- SciCodeLlama 4 Maverick by 10.431.721.3+10.4
- Terminal-Bench HardLlama 4 Maverick by 5.36.81.5+5.3
- Terminal-Bench 2.1Llama 4 Maverick by 4.27.93.7+4.2
AgentsEven
- τ-Bench V3 · Bankingtie3.73.3tie
- GDPValtie00tie
- Terminal-Bench 4.0tie00tie
ReasoningLlama 4 Maverick
- GPQA DiamondLlama 4 Maverick by 8.467.158.7+8.4
- Humanity's Last ExamLlama 4 Maverick by 1.14.93.8+1.1
- ARC-AGI-2tie00tie
- CritPttie00tie
FactsEven
- AA-Omniscience · AccuracyLlama 4 Maverick by 9.724.915.2+9.7
- AA-Omniscience · Non-hallucinationLlama 4 Scout by 9.511.120.6+9.5
Images and chartsLlama 4 Maverick
Following instructionsLlama 4 Maverick
- IFBenchLlama 4 Maverick by 3.54339.5+3.5
Other results12 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- LiveCodeBenchLlama 4 Maverick by 10.643.432.8+10.6
- AA-OmniscienceLlama 4 Maverick by 10.3-41.9-52.1+10.3
- MBPPLlama 4 Maverick by 9.877.667.8+9.8
- Artificial Analysis Coding IndexLlama 4 Maverick by 8.116.38.2+8.1
- MMLU-ProLlama 4 Maverick by 6.280.574.3+6.2
- image_reasoning_mmmuLlama 4 Maverick by 473.469.4+4
- image_reasoning_mathvistaLlama 4 Maverick by 373.770.7+3
- τ²-Bench Telecom (AA run)Llama 4 Maverick by 2.317.815.5+2.3
- AA IntelligenceLlama 4 Maverick by 1.9108.1+1.9
- MGSMLlama 4 Maverick by 1.792.390.6+1.7
- ChartQALlama 4 Maverick by 1.29088.8+1.2
- DocVQAtie94.494.4tie
Questions people ask
Which is better, Llama 4 Maverick or Llama 4 Scout?
Llama 4 Maverick wins five of the seven areas where both have results: coding, reasoning, images and charts, long documents and following instructions. Llama 4 Scout wins none, but costs 22% less. They are level on agents and facts.
Which is better for coding?
Llama 4 Maverick. It wins 4 of the 4 coding tests both models report; Llama 4 Scout wins none.
Which is cheaper?
Llama 4 Maverick costs $0.25 per million input tokens and $0.87 per million output tokens; Llama 4 Scout costs $0.19 and $0.68. That makes Llama 4 Scout about 22% cheaper for the same work.
How do you compare the two?
We use the 30 benchmark tests both models have published scores on. The verdict counts the 18 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 12 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.