Qwen3.6 27B: benchmark scores, price and ranking

Qwen3.6 27B from Alibaba ranks #615 of 743 LLMs on the LLMs Tiger aggregate benchmark score (44.9), based on 19 benchmark results captured up to 2026-10-13.

Organisation

Alibaba

Type

Open-weight / local model

Open weights

Yes

Released

2026-04-22 (source: epoch)

Cheapest API price

$0.150 in / $0.500 out per million tokens (via libertai, observed 2026-10-13T19:47Z)

Aggregate rank

#615 of 743 (score 44.9)

Relative to other models, Qwen3.6 27B is strongest on MMLU-Pro (#9 of 136) and weakest on LiveBench Reasoning (#17 of 17).

Across the 18 benchmarks where at least ten models are scored, Qwen3.6 27B is in the top 10% on 1 and in the top half on 7.

Its cheapest listed input price, $0.150 per million tokens, is lower than that of 72% of the 427 models with a listed price.

Qwen3.6 27B benchmark results

About Qwen3.6 27B

Qwen3.6 27B is an open-weight model that can run on your own hardware from Alibaba. It was released on 2026-04-22, according to Epoch AI. The cheapest API listing we track charges $0.150 per million input tokens and $0.500 per million output tokens, through libertai. We hold 19 benchmark results for it across 4 areas, and 1 is in the top tenth of its board.

Where Qwen3.6 27B stands

On our aggregate score, Qwen3.6 27B is #615 of 743 ranked models, in the lower half. Its score is 44.9; the scale is centred on 50, so it sits 5.1 points below the typical model. That beats 17% of the ranked field. Among the 74 ranked Alibaba models it is #67. The next one up is Qwen3-30B-A3B at 44.9. The next one down is Qwen3.5 9B at 44.9.

Qwen3.6 27B, benchmark by benchmark

On MMLU-Pro, Qwen3.6 27B scores 86.2% and ranks #9 of 136. Only 8 models score higher, led by Darwin-180B-RSI at 88.1%, which is 1.9 points ahead. It covers 57-subject knowledge + reasoning. The board median is 55.3%, so Qwen3.6 27B is 30.9 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.

On Epoch ECI, Qwen3.6 27B reaches 146. It ranks #21 of 204, ahead of 90% of the field, and 10 points behind Qwen 3.8 Max at 156. The results either side are Qwen3.5 397B-A17B (147) and Grok 4 (146). It covers Epoch Capabilities Index: Epoch's own aggregate of ~40 benchmarks. The board median is 127, so Qwen3.6 27B is 19 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Qwen3.6 27B manages 87.8% on GPQA Diamond, good for #54 of 275. GPT-6 Astra (max) leads at 95.8%, 8.0 points further up, and 80% of models sit below. The results either side are GLM-4.7-FP8 (87.9%) and GPT-5.1 (high) (87.6%). It covers expert-level science questions. The board median is 71.7%, so Qwen3.6 27B is 16.1 points above it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.

OTIS Mock AIME puts Qwen3.6 27B at 91.1%, #42 of 189. That beats 78% of the models tested, but 8.9 points short of Claude Fable 5.1 (max) (100.0%). The results either side are Qwen 3.6 Max (Preview) (91.1%) and Inkling Small (xhigh) (90.0%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so Qwen3.6 27B is 33.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On Chess Puzzles, Qwen3.6 27B reaches 22.0%. It ranks #39 of 122, ahead of 68% of the field, and 50.0 points behind GPT-6 Astra (max) at 72.0%. The results either side are Qwen3.5 Plus (22.0%) and GLM-5.3 (max) (21.0%). It covers chess tactics puzzles. The board median is 12.5%, so Qwen3.6 27B is 9.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Qwen3.6 27B manages 78.1% on DTBench, good for #59 of 146. Claude Opus 5.5 (max) leads at 98.9%, 20.8 points further up, and 60% of models sit below. The results either side are Qwen3-235B-A22B-Instruct-2507 (78.4%) and o4-mini (high) (77.6%). It covers dynamic tool-use benchmark. The board median is 68.8%, so Qwen3.6 27B is 9.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

LiveBench Language is a weak spot. Qwen3.6 27B scores 63.3%, #16 of 17, behind 88% of the models measured. The leader, GPT-5.5 (xhigh), reaches 87.4%. It covers LiveBench release 2026_06_25. The board median is 77.9%, so Qwen3.6 27B is 14.6 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

LiveBench Agentic Coding is a weak spot. Qwen3.6 27B scores 39.3%, #17 of 17, behind 94% of the models measured. The leader, claude-opus-4-8-xhigh-effort, reaches 56.1%. It covers LiveBench release 2026_06_25. The board median is 45.8%, so Qwen3.6 27B is 6.5 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

LiveBench IF is a weak spot. Qwen3.6 27B scores 53.2%, #17 of 17, behind 94% of the models measured. The leader, gemini-3.1-pro-preview-high, reaches 79.1%. It covers LiveBench release 2026_06_25. The board median is 65.2%, so Qwen3.6 27B is 12.0 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

LiveBench Reasoning is a weak spot. Qwen3.6 27B scores 70.3%, #17 of 17, behind 94% of the models measured. The leader, claude-opus-4-8-xhigh-effort, reaches 89.7%. It covers LiveBench release 2026_06_25. The board median is 81.1%, so Qwen3.6 27B is 10.8 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

The full table above also lists the other 9 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: weak, ahead of 18% of the field on average across 3 benchmarks. Its best board here is SciCode (#74 of 106) and its weakest is LiveBench Agentic Coding (#17 of 17). Reasoning, maths and science: uneven, ahead of 49% of the field on average across 7 benchmarks. Its best board here is MMLU-Pro (#9 of 136) and its weakest is LiveBench Reasoning (#17 of 17). General knowledge and chat: uneven, ahead of 37% of the field on average across 7 benchmarks. Its best board here is Epoch ECI (#21 of 204) and its weakest is LiveBench IF (#17 of 17).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so Qwen3.6 27B 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 Phi-3-medium-128k-instruct ($0.170) and Llama 4 Maverick (FP8) ($0.188).

Release history

The Alibaba model released before it was Qwen 3.6 Flash (2026-04-16) on 2026-04-16, which scores 47.2. The next one was qwen3.7-max-20260517 on 2026-05-17, at 47.2.

Reading these numbers

The 19 scores on this page come from 5 sources: Epoch AI Benchmarking Hub (9), LiveBench 2026_06_25 official CSV (7), HuggingFace GPQA Diamond leaderboard (1), HuggingFace MMLU-Pro leaderboard (1). 7 of them come from an official leaderboard and 12 from datasets or third-party runs.

Models ranked nearby

• #612 Qwen3-30B-A3B (44.9)
• #613 aya-expanse-8b (44.9)
• #614 Mercury 2 (44.9)
• #616 Nova 2.0 Pro Preview (unknown) (44.9)
• #617 internlm2-math-plus-7b (44.9)
• #618 Qwen3.5 9B (44.9)

More from Alibaba

• Qwen 3.6 Max (Preview) (#25, 54.1)
• Qwen3.8 Max (xhigh) (#37, 53.2)
• Qwen3.7 Max (#71, 52.3)
• ollama/qwen2.5-coder:32b (#93, 51.9)
• Qwen3 235B A22B diff, no think, Alibaba API (#122, 51.5)
• Qwen2.5-Coder-14B-Instruct (#135, 51.3)

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