qwen3-4b-instruct-2507: benchmark scores, price and ranking

qwen3-4b-instruct-2507 from Qwen ranks #333 of 743 LLMs on the LLMs Tiger aggregate benchmark score (48.7), based on 4 benchmark results captured up to 2026-10-13.

Organisation: Qwen
Type: Open-weight / local model
Open weights: Yes
Released: 2025-08-06 (source: epoch)
Aggregate rank: #333 of 743 (score 48.7)

Relative to other models, qwen3-4b-instruct-2507 is strongest on MMLU-Pro (#41 of 136) and weakest on Chess Puzzles (#86 of 122).

Across the 4 benchmarks where at least ten models are scored, qwen3-4b-instruct-2507 is in the top 10% on 0 and in the top half on 1.

qwen3-4b-instruct-2507 benchmark results

qwen3-4b-instruct-2507 benchmark scores, ranked against all models measured on each benchmark
Benchmark	Score	Rank	Source	Captured
MMLU-Pro	69.6%	#41 of 136	HuggingFace MMLU-Pro leaderboard	2026-10-13
OTIS Mock AIME	52.2%	#104 of 189	Epoch AI Benchmarking Hub	2026-10-13
GPQA Diamond	62.0%	#164 of 275	HuggingFace GPQA Diamond leaderboard	2026-10-13
Chess Puzzles	4.0%	#86 of 122	Epoch AI Benchmarking Hub	2026-10-13

About qwen3-4b-instruct-2507

qwen3-4b-instruct-2507 is an open-weight model that can run on your own hardware from Qwen. It was released on 2025-08-06, according to Epoch AI. No API price is listed for it yet. We hold 4 benchmark results for it across 2 areas, and none is in the top tenth of its board.

Where qwen3-4b-instruct-2507 stands

On our aggregate score, qwen3-4b-instruct-2507 is #333 of 743 ranked models, around the middle of the table. Its score is 48.7; the scale is centred on 50, so it sits 1.3 points below the typical model. That beats 55% of the ranked field. Among the 23 ranked Qwen models it is #9. The next one up is Qwen3-30B-A3B-Base at 48.7. The next one down is SWE-agent-LM-32B at 48.5.

qwen3-4b-instruct-2507, benchmark by benchmark

MMLU-Pro puts qwen3-4b-instruct-2507 at 69.6%, #41 of 136. That beats 70% of the models tested, but 18.5 points short of Darwin-180B-RSI (88.1%). The results either side are phi-4 (70.4%) and ERNIE-4.5-300B-A47B-Base-PT (69.5%). It covers 57-subject knowledge + reasoning. The board median is 55.3%, so qwen3-4b-instruct-2507 is 14.3 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.

qwen3-4b-instruct-2507 lands below the middle on OTIS Mock AIME: 52.2%, #104 of 189, with 54% of the field scoring higher. Claude Fable 5.1 (max) leads at 100.0%; the closest result above is DeepSeek-R1 at 53.3%. It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so qwen3-4b-instruct-2507 is 5.6 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

qwen3-4b-instruct-2507 lands below the middle on GPQA Diamond: 62.0%, #164 of 275, with 59% of the field scoring higher. GPT-6 Astra (max) leads at 95.8%; the closest result above is o1-mini (high) at 62.4%. It covers expert-level science questions. The board median is 71.7%, so qwen3-4b-instruct-2507 is 9.7 points below it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.

qwen3-4b-instruct-2507 lands below the middle on Chess Puzzles: 4.0%, #86 of 122, with 70% of the field scoring higher. GPT-6 Astra (max) leads at 72.0%; the closest result above is Qwen3-30B-A3B at 4.0%. It covers chess tactics puzzles. The board median is 12.5%, so qwen3-4b-instruct-2507 is 8.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Strengths and weaknesses

Reasoning, maths and science: solid, ahead of 52% of the field on average across 3 benchmarks. Its best board here is MMLU-Pro (#41 of 136) and its weakest is GPQA Diamond (#164 of 275).

Release history

The Qwen model released before it was Qwen3-Coder-30B-A3B-Instruct on 2025-07-31, which scores 52.7. The next one was Qwen3.5-122B-A10B on 2026-02-24, at 53.0.

Reading these numbers

The 4 scores on this page come from 3 sources: Epoch AI Benchmarking Hub (2), HuggingFace GPQA Diamond leaderboard (1), HuggingFace MMLU-Pro leaderboard (1). No official leaderboard supplied any of them, so treat them as third-party measurements. With 4 results the aggregate rank can move a lot when one new score arrives.

Models ranked nearby

#330 MiniMax-M2 (48.7)
#331 gemini-1.5-flash-exp-0827 (48.7)
#332 grok-4.20-beta-0309-reasoning (48.7)
#334 DeepSWE-Preview (48.6)
#335 chatgpt-4o-latest (2025-02-15) (48.6)
#336 Nemotron-H-56B-Base-8K (48.6)