Llama-3.2-90B-Vision-Instruct: benchmark scores, price and ranking
Llama-3.2-90B-Vision-Instruct from Meta AI ranks #553 of 743 LLMs on the LLMs Tiger aggregate benchmark score (45.9), based on 5 benchmark results captured up to 2026-10-13.
- Organisation
- Meta AI
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2024-09-24 (source: epoch)
- Cheapest API price
- $2.00 in / $2.00 out per million tokens (via watsonx, observed 2026-10-13T20:47Z)
- Aggregate rank
- #553 of 743 (score 45.9)
Relative to other models, Llama-3.2-90B-Vision-Instruct is strongest on GSM8K (#4 of 17) and weakest on EnigmaEval (#24 of 24).
Across the 5 benchmarks where at least ten models are scored, Llama-3.2-90B-Vision-Instruct is in the top 10% on 0 and in the top half on 1.
Its cheapest listed input price, $2.00 per million tokens, is lower than that of 18% of the 427 models with a listed price.
Llama-3.2-90B-Vision-Instruct benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| GSM8K | 93.1% | #4 of 17 | HuggingFace GSM8K leaderboard | 2026-10-13 |
| BALROG | 27.3% | #18 of 34 | Epoch AI Benchmarking Hub | 2026-10-13 |
| GPQA Diamond | 46.1% | #205 of 275 | HuggingFace GPQA Diamond leaderboard | 2026-10-13 |
| OTIS Mock AIME | 2.6% | #167 of 189 | Epoch AI Benchmarking Hub | 2026-10-13 |
| EnigmaEval | 0.4% | #24 of 24 | Epoch AI Benchmarking Hub | 2026-10-13 |
About Llama-3.2-90B-Vision-Instruct
Llama-3.2-90B-Vision-Instruct is an open-weight model that can run on your own hardware from Meta AI. It was released on 2024-09-24, according to Epoch AI. The cheapest API listing we track charges $2.00 per million input tokens and $2.00 per million output tokens, through watsonx. We hold 5 benchmark results for it across 3 areas, and none is in the top tenth of its board.
Where Llama-3.2-90B-Vision-Instruct stands
On our aggregate score, Llama-3.2-90B-Vision-Instruct is #553 of 743 ranked models, in the lower half. Its score is 45.9; the scale is centred on 50, so it sits 4.1 points below the typical model. That beats 26% of the ranked field. Among the 28 ranked Meta AI models it is #13. The next one up is Llama-4-Maverick-17B-128E-Instruct at 46.9. The next one down is Llama 4 Maverick (FP8) at 45.6.
Llama-3.2-90B-Vision-Instruct, benchmark by benchmark
GSM8K puts Llama-3.2-90B-Vision-Instruct at 93.1%, #4 of 17. That beats 76% of the models tested, but 6.5 points short of mimo-v2.5-pro (99.6%). The results either side are granite-4.1-30b (94.2%) and granite-4.1-8b (92.5%). It covers grade-school math word problems. The board median is 86.9%, so Llama-3.2-90B-Vision-Instruct is 6.2 points above it. Source: HuggingFace GSM8K leaderboard, captured 2026-10-13.
Llama-3.2-90B-Vision-Instruct lands below the middle on BALROG: 27.3%, #18 of 34, with 50% of the field scoring higher. GPT-6 Astra (max) leads at 68.3%; the closest result above is Llama-3.1-70B-Instruct at 27.9%. It covers long-horizon game-playing agents. The board median is 27.6%, so Llama-3.2-90B-Vision-Instruct is 0.3 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Llama-3.2-90B-Vision-Instruct lands below the middle on GPQA Diamond: 46.1%, #205 of 275, with 74% of the field scoring higher. GPT-6 Astra (max) leads at 95.8%; the closest result above is qwen2.5-32b-instruct at 46.1%. It covers expert-level science questions. The board median is 71.7%, so Llama-3.2-90B-Vision-Instruct is 25.6 points below it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.
OTIS Mock AIME is a weak spot. Llama-3.2-90B-Vision-Instruct scores 2.6%, #167 of 189, behind 88% of the models measured. The leader, Claude Fable 5.1 (max), reaches 100.0%. It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so Llama-3.2-90B-Vision-Instruct is 55.1 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
EnigmaEval is a weak spot. Llama-3.2-90B-Vision-Instruct scores 0.4%, #24 of 24, behind 96% of the models measured. The leader, Gemini 3 Pro Preview, reaches 18.2%. It covers multi-step puzzle hunts. The board median is 2.9%, so Llama-3.2-90B-Vision-Instruct is 2.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Strengths and weaknesses
Reasoning, maths and science: weak, ahead of 12% of the field on average across 3 benchmarks. Its best board here is GPQA Diamond (#205 of 275) and its weakest is EnigmaEval (#24 of 24).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so Llama-3.2-90B-Vision-Instruct costs more than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on GPT-4o ($2.50) and Command R+ (08-24) ($2.50).
Release history
The Meta AI model released before it was Llama-3.2-3B-Instruct on 2024-09-24, which scores 47.2. The next one was Llama-3.3-70B-Instruct on 2024-12-06, at 40.7.
Reading these numbers
The 5 scores on this page come from 3 sources: Epoch AI Benchmarking Hub (3), HuggingFace GPQA Diamond leaderboard (1), HuggingFace GSM8K leaderboard (1). No official leaderboard supplied any of them, so treat them as third-party measurements. With 5 results the aggregate rank can move a lot when one new score arrives.
Models ranked nearby
- #550 Mixtral-8x7B-v0.1 (45.9)
- #551 dbrx-instruct (45.9)
- #552 MAmmoTH2-7B-Plus (45.9)
- #554 Laguna XS.2 (45.9)
- #555 codeqwen:7b-chat-v1.5-q8_0 (45.8)
- #556 claude-instant-1.2 (45.8)
More from Meta AI
- Muse Spark 1.3 (max) (#18, 55.1)
- Muse Spark (#29, 53.7)
- Muse Spark 1.1 (xhigh) (#66, 52.4)
- Muse Spark 1.2 (xhigh) (#113, 51.5)
- Llama 3.1-405B (#142, 51.2)
- Muse Wonder (max) (#209, 50.1)
Compare it on the live leaderboard → · How the aggregate score works