What are Qwen3.5-27B's benchmark scores?

As of 25 Oct 2026, Qwen3.5-27B (thinking effort) scores 54.7 ±8.0 on the BenchLeader Index, strongest in instruction following (74) and weakest in knowledge (42). That rests on 7 independent benchmark results, and every score on this page links to its publisher.

How does Qwen3.5-27B rank?

Qwen3.5-27B (thinking effort) is #246 of 785 ranked configurations on 25 Oct 2026. 405 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.

How much does Qwen3.5-27B cost per 1M tokens?

As of 25 Oct 2026, Qwen3.5-27B costs $0.300 per 1M input tokens and $2.40 per 1M output tokens, $0.825 blended (three input tokens per output token).

When was Qwen3.5-27B released?

Alibaba released Qwen3.5-27B on 24 Feb 2026. BenchLeader has results for it at 3 reasoning-effort settings.

Alibaba Open weights: Qwen/Qwen3.5-27B Compare with another model

Qwen3.5-27B

Reasoning effort

Qwen3.5-27B is an Alibaba open-weights model, released 24 Feb 2026. Its best configuration (thinking reasoning effort) ranks #246 of 785 on the BenchLeader Index at 54.7 ±8.0, in the upper half. The ± is the point: 405 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (74) and lowest in knowledge (42). At $0.825 per million tokens blended it is mid-priced. Output speed of 78 tokens per second puts it slower than most, with a first answer in 31.2 s. It has been measured at 3 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 24 Feb 2026.stale: no new result in six months

Blended price: $0.825/M; $0.300 in · $2.40 out
Output speed: 78 tok/s; measured by Artificial Analysis
First answer: 31 s; first token 1.92 s
Context: 262k
Full answer: 38 s; median, reasoning included
Cost per run: –
Released: 24 Feb 2026

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index55
Reasoning42
Agents & tools62
Knowledge42
Instruction following74
Multimodal58
Long context64
Composite55

Versions

Alibaba has shipped 3 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
Qwen3.8 27Bxhigh	14 Aug 2026	56.3	#207
Qwen3.6 27Bthinking	22 Apr 2026	53.9	#266
Qwen3.5-27Bthinkingthis page	24 Feb 2026	54.7	#246

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
no reasoning	50.2	#378	81 tok/s	5.55 s	$0.0008	Agents & tools 62 · Composite 51 · Instruction following 49 · Knowledge 40 · Long context 57 · Multimodal 53 · Reasoning 41
thinkingbest	54.7	#246	78 tok/s	31 s	$0.0008	Agents & tools 62 · Composite 55 · Instruction following 74 · Knowledge 42 · Long context 64 · Multimodal 58 · Reasoning 42
not stated	52.3	#320	42 tok/s	1.92 s	$0.0008	Coding 45 · Human preference 59 · Maths 46 · Multimodal 57 · Reasoning 59

How its index has moved

8 Sept 2026 to 23 Oct 2026

thinking
not stated
no reasoning

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Metadata

Benchmark	thinking	Source	Trend
AA Openness Indexnot in index	38.9#34	Artificial Analysis

Reasoning

Benchmark	no reasoning	thinking	not stated	Source
Arena Hard Promptseffort not stated	–	–	1428#138	Arena (LMArena)
GPQA Diamond (AA)not in index	84.2%#145	85.8%#119	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	13.9%#262	23.9%#190	–	Artificial Analysis
CritPt	0.3%#271	0.8%#252	–	Artificial Analysis
LMCAnot in indexeffort not stated	–	–	34.0%#129	Epoch AI Benchmarking Hub
DTBenchnot in indexeffort not stated	–	–	82.4%#120	Epoch AI Benchmarking Hub

Coding

Benchmark	not stated	Source
WeirdML	39.5%#105	WeirdML
Arena Coding	1452#145	Arena (LMArena)
Arena WebDev	1359#107	Arena (LMArena)
ALE-Benchnot in index	349.4#111	Epoch AI Benchmarking Hub

Agents & tools

Benchmark	no reasoning	thinking	not stated	Source
Terminal-Bench Hard	31.8%#101	32.6%#97	–	Artificial Analysis
τ²-Bench Telecom (AA)not in index	87.1%#80	93.9%#38	–	Artificial Analysis
ITBench SRE (AA)not in index	–	35.5%#26	–	Artificial Analysis
Vending-Bench 2not in indexeffort not stated	–	–	202.0#61	Andon Labs

Maths

Benchmark	not stated	Source
AIME 2026possible contamination	91.7%#24	MathArena
HMMT February 2026possible contamination	81.1%#26	MathArena
MathArena Apexpossible contamination	2.1%#29	MathArena
Arena Maths	1423#116	Arena (LMArena)

Knowledge

Benchmark	no reasoning	thinking	Source
AA-Omniscience	-47.7#368	-44.0#344	Artificial Analysis
AA-Omniscience: accuracynot in index	15.7%#405	20.7%#295	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	24.9%#244	18.5%#297	Artificial Analysis
Global-MMLU-Lite (AA)not in index	–	86.8%#22	Artificial Analysis

Instruction following

Benchmark	no reasoning	thinking	not stated	Source	Trend
IFBench	46.9%#192	75.6%#29	–	Artificial Analysis
Arena Instruction Followingnot in indexeffort not stated	–	–	1401#137	Arena (LMArena)

Human preference

Benchmark	not stated	Source
Arena Text	1409#142	Arena (LMArena)
Arena Creative Writingnot in index	1358#161	Arena (LMArena)
Arena Multi-turnnot in index	1414#134	Arena (LMArena)
Arena Longer Queriesnot in index	1423#127	Arena (LMArena)

Multimodal

Benchmark	no reasoning	thinking	not stated	Source	Trend
Arena Visioneffort not stated	–	–	1241#71	Arena (LMArena)
MMMU-Pro	70.0%#150	75.0%#104	–	Artificial Analysis

Long context

Benchmark	no reasoning	thinking	Source	Trend
AA-LCR	64.0%#260	77.7%#128	Artificial Analysis

Composite

Benchmark	no reasoning	thinking	Source	Trend
AA Intelligence Index v4.3.2	19.4#242	22.9#193	Artificial Analysis

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
Alibaba Cloud Int.	76 tok/s	0.74 s	$0.195	$1.56	262k	–
DeepInfra	63 tok/s	0.70 s	$0.260	$2.60	262k	fp8
AtlasCloud	58 tok/s	0.80 s	$0.270	$2.16	262k	fp8
NovitaAI	57 tok/s	1.47 s	$0.300	$2.40	262k	bf16
SiliconFlow	34 tok/s	1.29 s	$0.250	$2.00	262k	fp8
Phala	26 tok/s	1.73 s	$0.300	$2.40	262k	–

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0008	–	35.0 s
Summarize a 30-page report	12,000 / 600	$0.0050	–	38.9 s
Code edit	6,000 / 1,500	$0.0054	–	50.4 s
Agentic coding session	60,000 / 4,000	$0.028	–	1.4 min
Structured extraction	2,000 / 200	$0.0011	–	33.7 s

New this month

Questions people ask

What are Qwen3.5-27B's benchmark scores?: As of 25 Oct 2026, Qwen3.5-27B (thinking effort) scores 54.7 ±8.0 on the BenchLeader Index, strongest in instruction following (74) and weakest in knowledge (42). That rests on 7 independent benchmark results, and every score on this page links to its publisher.
How does Qwen3.5-27B rank?: Qwen3.5-27B (thinking effort) is #246 of 785 ranked configurations on 25 Oct 2026. 405 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
How much does Qwen3.5-27B cost per 1M tokens?: As of 25 Oct 2026, Qwen3.5-27B costs $0.300 per 1M input tokens and $2.40 per 1M output tokens, $0.825 blended (three input tokens per output token).
When was Qwen3.5-27B released?: Alibaba released Qwen3.5-27B on 24 Feb 2026. BenchLeader has results for it at 3 reasoning-effort settings.

Data as of 25 Oct 2026. Compare these configurations.

Cite as: BenchLeader, “Qwen3.5-27B: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/qwen3-5-27b.