LLMs TigerLive LLM benchmarks

Qwen3.6 35B-A3B: benchmark scores, price and ranking

Qwen3.6 35B-A3B from Alibaba ranks #535 of 743 LLMs on the LLMs Tiger aggregate benchmark score (46.2), based on 15 benchmark results captured up to 2026-10-13.

Organisation
Alibaba
Type
Open-weight / local model
Open weights
Yes
Released
2026-04-14 (source: epoch)
Cheapest API price
$0.050 in / $0.400 out per million tokens (via sail, observed 2026-10-13T19:47Z)
Aggregate rank
#535 of 743 (score 46.2)

Relative to other models, Qwen3.6 35B-A3B is strongest on MMLU-Pro (#13 of 136) and weakest on Terminal-Bench 2.0 (#25 of 28).

Across the 14 benchmarks where at least ten models are scored, Qwen3.6 35B-A3B is in the top 10% on 1 and in the top half on 6.

Its cheapest listed input price, $0.050 per million tokens, is lower than that of 91% of the 427 models with a listed price.

Qwen3.6 35B-A3B benchmark results

Qwen3.6 35B-A3B benchmark scores, ranked against all models measured on each benchmark
BenchmarkScoreRankSourceCaptured
MMLU-Pro85.2%#13 of 136HuggingFace MMLU-Pro leaderboard2026-10-13
GPQA Diamond86.0%#72 of 275HuggingFace GPQA Diamond leaderboard2026-10-13
Chess Puzzles26.0%#32 of 122Epoch AI Benchmarking Hub2026-10-13
OTIS Mock AIME86.7%#55 of 189Epoch AI Benchmarking Hub2026-10-13
LMCA29.7%#50 of 108Epoch AI Benchmarking Hub2026-10-13
DTBench73.9%#69 of 146Epoch AI Benchmarking Hub2026-10-13
Mystery Games22.0%#31 of 59Epoch AI Benchmarking Hub2026-10-13
Surface Evolver44.4%#9 of 13Epoch AI Benchmarking Hub2026-10-13
CritPt0.3%#84 of 115Epoch AI Benchmarking Hub2026-10-13
SciCode35.8%#79 of 106Epoch AI Benchmarking Hub2026-10-13
WeirdML34.5%#82 of 109Epoch AI Benchmarking Hub2026-10-13
FrontierMath T1-320.4%#54 of 68Epoch AI Benchmarking Hub2026-10-13
SWE-rebench24.7%#6 of 7SWE-rebench leaderboard2026-10-13
WebDev Arena1,250#58 of 67Epoch AI Benchmarking Hub2026-10-13
Terminal-Bench 2.023.0%#25 of 28Epoch AI Benchmarking Hub2026-10-13

About Qwen3.6 35B-A3B

Qwen3.6 35B-A3B is an open-weight model that can run on your own hardware from Alibaba. It was released on 2026-04-14, according to Epoch AI. The cheapest API listing we track charges $0.050 per million input tokens and $0.400 per million output tokens, through sail. We hold 15 benchmark results for it across 4 areas, and 1 is in the top tenth of its board.

Where Qwen3.6 35B-A3B stands

On our aggregate score, Qwen3.6 35B-A3B is #535 of 743 ranked models, in the lower half. Its score is 46.2; the scale is centred on 50, so it sits 3.8 points below the typical model. That beats 28% of the ranked field. Among the 74 ranked Alibaba models it is #57. The next one up is qwen1.5-110b-chat at 46.3. The next one down is qwen1.5-32b-chat at 45.7.

Qwen3.6 35B-A3B, benchmark by benchmark

On MMLU-Pro, Qwen3.6 35B-A3B scores 85.2% and ranks #13 of 136. Only 12 models score higher, led by Darwin-180B-RSI at 88.1%, which is 2.9 points ahead. It covers 57-subject knowledge + reasoning. The board median is 55.3%, so Qwen3.6 35B-A3B is 29.9 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.

On GPQA Diamond, Qwen3.6 35B-A3B reaches 86.0%. It ranks #72 of 275, ahead of 74% of the field, and 9.8 points behind GPT-6 Astra (max) at 95.8%. The results either side are GLM-5 (86.0%) and Darwin-31B-Opus (85.9%). It covers expert-level science questions. The board median is 71.7%, so Qwen3.6 35B-A3B is 14.3 points above it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.

Qwen3.6 35B-A3B manages 26.0% on Chess Puzzles, good for #32 of 122. GPT-6 Astra (max) leads at 72.0%, 46.0 points further up, and 74% of models sit below. The results either side are o4-mini (high) (26.0%) and grok-4.3 (high) (25.0%). It covers chess tactics puzzles. The board median is 12.5%, so Qwen3.6 35B-A3B is 13.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

OTIS Mock AIME puts Qwen3.6 35B-A3B at 86.7%, #55 of 189. That beats 71% of the models tested, but 13.3 points short of Claude Fable 5.1 (max) (100.0%). The results either side are Qwen3.5 Plus (86.7%) and Qwen3.7 Flash (86.7%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so Qwen3.6 35B-A3B is 28.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On LMCA, Qwen3.6 35B-A3B reaches 29.7%. It ranks #50 of 108, ahead of 54% of the field, and 38.5 points behind Claude Opus 5.5 (max) at 68.2%. The results either side are Qwen 3.6 Flash (2026-04-16) (31.0%) and Gemma 4 26B A4B (29.7%). It covers language-model capability assessment. The board median is 29.2%, so Qwen3.6 35B-A3B is 0.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Qwen3.6 35B-A3B manages 73.9% on DTBench, good for #69 of 146. Claude Opus 5.5 (max) leads at 98.9%, 25.1 points further up, and 53% of models sit below. The results either side are o1 (high) (74.7%) and Qwen3.5 9B (71.2%). It covers dynamic tool-use benchmark. The board median is 68.8%, so Qwen3.6 35B-A3B is 5.1 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Qwen3.6 35B-A3B lands below the middle on WeirdML: 34.5%, #82 of 109, with 74% of the field scoring higher. GPT-6 Astra (max) leads at 93.3%; the closest result above is grok-code-fast-1 at 35.1%. It covers unusual machine-learning tasks. The board median is 41.2%, so Qwen3.6 35B-A3B is 6.7 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Qwen3.6 35B-A3B lands below the middle on FrontierMath T1-3: 20.4%, #54 of 68, with 78% of the field scoring higher. GPT-6 Astra (max) leads at 93.7%; the closest result above is Qwen 3.6 Flash (2026-04-16) at 22.5%. It covers research-level maths, tiers 1-3. The board median is 55.6%, so Qwen3.6 35B-A3B is 35.3 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

WebDev Arena is a weak spot. Qwen3.6 35B-A3B scores 1,250, #58 of 67, behind 85% of the models measured. The leader, Claude Opus 5.5 (max), reaches 1,820. It covers building web apps, crowd-voted (Elo). The board median is 1,433, so Qwen3.6 35B-A3B is 183 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Terminal-Bench 2.0 is a weak spot. Qwen3.6 35B-A3B scores 23.0%, #25 of 28, behind 86% of the models measured. The leader, Gemini 3 Pro Preview, reaches 69.4%. It covers real terminal tasks (agent + model). The board median is 37.3%, so Qwen3.6 35B-A3B is 14.3 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

The full table above also lists the other 5 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: weak, ahead of 17% of the field on average across 3 benchmarks. Its best board here is SciCode (#79 of 106) and its weakest is Terminal-Bench 2.0 (#25 of 28). Reasoning, maths and science: solid, ahead of 57% of the field on average across 5 benchmarks. Its best board here is MMLU-Pro (#13 of 136) and its weakest is FrontierMath T1-3 (#54 of 68). General knowledge and chat: uneven, ahead of 46% of the field on average across 5 benchmarks. Its best board here is Chess Puzzles (#32 of 122) and its weakest is WeirdML (#82 of 109).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so Qwen3.6 35B-A3B costs less 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 amazon.nova-lite-v1:0 ($0.060) and qwen2.5-32b-instruct ($0.060).

Release history

The Alibaba model released before it was qwen3.6-plus on 2026-04-02, which scores 46.6. The next one was Qwen 3.6 Flash (2026-04-16) on 2026-04-16, at 47.2.

Reading these numbers

The 15 scores on this page come from 4 sources: Epoch AI Benchmarking Hub (12), HuggingFace GPQA Diamond leaderboard (1), HuggingFace MMLU-Pro leaderboard (1), SWE-rebench leaderboard (1). No official leaderboard supplied any of them, so treat them as third-party measurements.

Models ranked nearby

More from Alibaba

Compare it on the live leaderboard → · How the aggregate score works