AlibabaQwen3.8 27B
Qwen3.8 27B is an Alibaba open-weights reasoning model, released 14 Aug 2026. Its best configuration (xhigh reasoning effort) ranks #211 of 785 on the BenchLeader Index at 56.2 ±6.6, in the upper half. The ± is the point: 327 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 composite (67) and lowest in agents & tools (47). At $1.13 per million tokens blended it is pricier than most ranked models. Output speed of 45 tokens per second puts it in the slowest quarter, with a first answer in 48.1 s. It has been measured at 5 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 20 Oct 2026.
- Blended price
- $1.13/M
- $0.500 in · $3.00 out
- Output speed
- 45 tok/s
- measured by Artificial Analysis
- First answer
- 48 s
- first token 0.83 s
- Context
- 262k
- Full answer
- 59 s
- median, reasoning included
- Cost per run
- $1.01
- one full Intelligence Index run
- Released
- 14 Aug 2026
- Overall index56
- Reasoning53
- Coding52
- Agents & tools47
- Knowledge56
- Multimodal59
- Long context66
- Composite67
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 27Bxhighthis page | 14 Aug 2026 | 56.2 | #211 |
| Qwen3.6 27Bthinking | 22 Apr 2026 | 53.9 | #266 |
| Qwen3.5-27Bthinking | 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 | 49.2 | #405 | 46 tok/s | 3.82 s | $0.0011 | Agents & tools 48 · Coding 35 · Composite 52 · Knowledge 58 · Long context 60 · Multimodal 53 · Reasoning 37 |
| low | 52.6 | #311 | 53 tok/s | 42 s | $0.0011 | Agents & tools 61 · Coding 40 · Composite 59 · Knowledge 50 · Long context 64 · Multimodal 57 · Reasoning 44 |
| medium | 53.8 | #270 | 51 tok/s | 44 s | $0.0011 | Agents & tools 77 · Coding 39 · Composite 60 · Knowledge 45 · Long context 65 · Multimodal 57 · Reasoning 42 |
| xhighbest | 56.2 | #211 | 45 tok/s | 48 s | $0.0011 | Agents & tools 47 · Coding 52 · Composite 67 · Knowledge 56 · Long context 66 · Multimodal 59 · Reasoning 53 |
| not stated | 54.6 | #249 | 54 tok/s | 0.83 s | $0.0011 | Agents & tools 44 · Coding 63 · Composite 45 · Human preference 62 · Maths 55 · Multimodal 61 · Reasoning 62 |
What more thinking costs
Turning the effort up buys index points and multiplies the bill. Here no reasoning costs 2.5× xhigh for -7.0 on the index — -5.4 points per doubling of spend. Compare that with other models.
How its index has moved
8 Sept 2026 to 24 Oct 2026- xhigh
- not stated
- medium
- low
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 | xhigh | Source | Trend |
|---|---|---|---|
| AA Openness Indexnot in index | 38.9#31 | Artificial Analysis |
Reasoning
| Benchmark | no reasoning | low | medium | xhigh | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|
| Arena Hard Promptseffort not stated | – | – | – | – | 1462#91 | Arena (LMArena) | |
| LiveBench 2.0 Reasoningnot in indexeffort not stated | – | – | – | – | 80.0%#40 | LiveBench | |
| LiveBench 1.0 Reasoningnot in indexeffort not stated | – | – | – | – | 80.0%#53 | LiveBench | |
| GPQA Diamond (AA)not in index | 81.8%#177 | 84.5%#137 | 84.5%#137 | 90.5%#59 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 12.1%#281 | 14.0%#261 | 14.1%#259 | 33.9%#130 | – | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | – | – | 88.9%#31 | – | Vals AI | |
| ARC-AGI-1 | 34.0%#201 | 69.2%#145 | 68.7%#146 | 87.5%#93 | – | ARC Prize | |
| ARC-AGI-2 | 1.5%#214 | 22.8%#141 | 13.2%#153 | 42.4%#116 | – | ARC Prize | |
| CritPt | 0.3%#271 | 0.0%#315 | 0.0%#315 | 5.4%#153 | – | Artificial Analysis | |
| LMCAnot in indexeffort not stated | – | – | – | – | 41.4%#104 | Epoch AI Benchmarking Hub | |
| DTBenchnot in indexeffort not stated | – | – | – | – | 88.0%#96 | Epoch AI Benchmarking Hub |
Coding
| Benchmark | no reasoning | low | medium | xhigh | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|
| SciCode | 36.2%#183 | 40.0%#162 | 39.0%#169 | 46.6%#132 | – | SciCode | |
| Arena Codingeffort not stated | – | – | – | – | 1499#81 | Arena (LMArena) | |
| Arena WebDeveffort not stated | – | – | – | – | 1591#29 | Arena (LMArena) | |
| LiveBench 2.0 Codingnot in indexeffort not stated | – | – | – | – | 86.8%#18 | LiveBench | |
| LiveBench 1.0 Codingnot in indexeffort not stated | – | – | – | – | 75.7%#50 | LiveBench | |
| SciCode (AA)not in index | 36.2%#196 | 40.0%#174 | 39.0%#181 | 46.6%#138 | – | Artificial Analysis | |
| LiveCodeBench | – | – | – | 84.0%#41 | – | Vals AI | |
| IOI | – | – | – | 39.1%#36 | – | Vals AI | |
| SWE-bench (Vals)not in index | – | – | – | 86.0%#15 | – | Vals AI | |
| Code Migrationnot in index | – | – | – | 14.2%#56 | – | Vals AI | |
| ProgramBenchnot in index | – | – | – | 0.0%#32 | – | Vals AI | |
| Vibe Code Bench v1.1not in index | – | – | – | 64.8%#48 | – | Vals AI | |
| Surface Evolver Benchnot in index | – | – | – | 45.0%#21 | – | Epoch AI Benchmarking Hub | |
| Terminal-Bench 4.0 (AA)not in index | 0.0%#152 | 2.5%#113 | 5.0%#102 | 5.6%#100 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 49.1%#122 | 67.4%#78 | 65.2%#82 | 79.8%#46 | – | Artificial Analysis | |
| FrontierCode 1.1 (AA)not in index | 7.9%#66 | 28.9%#54 | 34.7%#44 | 34.2%#45 | – | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | low | medium | xhigh | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|
| APEX-Agentseffort not stated | – | – | – | – | 47.5%#32 | Mercor | |
| Arena Agenteffort not stated | – | – | – | – | -3.8#37 | Arena (LMArena) | |
| LiveBench Long Agenticnot in indexeffort not stated | – | – | – | – | 27.9%#36 | LiveBench | |
| LiveBench Browsernot in indexeffort not stated | – | – | – | – | 32.0%#37 | LiveBench | |
| LiveBench Agentic Codingnot in indexeffort not stated | – | – | – | – | 61.4%#13 | LiveBench | |
| Terminal-Bench 2.1 (Vals) | – | – | – | 58.4%#40 | – | Vals AI | |
| GDPval-AA v2.1not in index | 28.7%#88 | 43.2%#55 | 44.0%#53 | 46.0%#49 | – | Artificial Analysis | |
| τ³-Banking (AA) | 20.0%#99 | 32.2%#58 | 47.4%#7 | 48.0%#6 | – | Artificial Analysis | |
| CyberBenchnot in index | – | – | – | 67.8%#28 | – | Vals AI | |
| Finance Agent v2not in index | – | – | – | 48.5%#49 | – | Vals AI | |
| Harvey's Legal Agent Benchmark v1not in index | – | – | – | 11.3%#15 | – | Vals AI | |
| Legal Research Benchnot in index | – | – | – | 36.1%#42 | – | Vals AI | |
| SkillsBenchnot in index | – | – | – | 38.1%#30 | – | Vals AI | |
| Tax Agent Benchnot in index | – | – | – | 30.9%#26 | – | Vals AI | |
| Terminal-Bench Sciencenot in index | – | – | – | 0.0%#38 | – | Vals AI | |
| AutomationBenchnot in index | 12.4%#151 | 30.1%#116 | 40.3%#99 | 48.2%#83 | – | Artificial Analysis | |
| GDP.pdf (AA)not in index | 5.2%#163 | 10.4%#129 | 12.8%#109 | 16.4%#90 | – | Artificial Analysis | |
| EnterpriseOps-Gymnot in index | – | – | – | 44.2%#16 | – | Artificial Analysis | |
| AA-Briefcase v1.1not in index | – | – | – | 1394#39 | – | Artificial Analysis | |
| Terminal-Bench-Science 0.1 (AA)not in index | 0.0%#71 | 0.5%#67 | 0.9%#63 | 0.5%#67 | – | Artificial Analysis | |
| CWE-Bench (AA) | – | – | – | 26.7%#20 | – | Artificial Analysis | |
| DeepSec-Bench (AA) | – | – | – | 11.5%#18 | – | Artificial Analysis | |
| CyberGym end-to-end (AA) | – | – | – | 0.0%#18 | – | Artificial Analysis |
Maths
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| ProofBench | 16.0%#56 | Vals AI | |
| LiveBench Mathematicsnot in index | 86.2%#52 | LiveBench | |
| Arena Maths | 1451#74 | Arena (LMArena) |
Knowledge
| Benchmark | no reasoning | low | medium | xhigh | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|
| LiveBench 2.0 Data Analysisnot in indexeffort not stated | – | – | – | – | 76.6%#22 | LiveBench | |
| LiveBench Researchnot in indexeffort not stated | – | – | – | – | 59.4%#44 | LiveBench | |
| LiveBench 1.0 Data Analysisnot in indexeffort not stated | – | – | – | – | 76.6%#31 | LiveBench | |
| AA-Omniscience | -8.0#177 | -26.7#266 | -36.1#306 | -10.0#191 | – | Artificial Analysis | |
| MMLU-Pro | – | – | – | 84.3%#62 | – | Vals AI | |
| LegalBench | – | – | – | 82.4%#70 | – | Vals AI | |
| TaxEval | – | – | – | 70.8%#82 | – | Vals AI | |
| Excel Modeling Benchmarknot in index | – | – | – | 59.7%#37 | – | Vals AI | |
| MedCodenot in index | – | – | – | 28.7%#95 | – | Vals AI | |
| MedScribenot in index | – | – | – | 83.8%#40 | – | Vals AI | |
| AA-Omniscience: accuracynot in index | 8.6%#511 | 17.1%#372 | 18.4%#338 | 15.6%#410 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 81.9%#11 | 47.1%#136 | 33.3%#201 | 69.7%#44 | – | Artificial Analysis |
Instruction following
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| LiveBench Languagenot in index | 74.3%#51 | LiveBench | |
| LiveBench Instruction Followingnot in index | 72.7%#19 | LiveBench | |
| Arena Instruction Followingnot in index | 1424#106 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1437#99 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1365#152 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1440#102 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1445#97 | Arena (LMArena) |
Multimodal
| Benchmark | no reasoning | low | medium | xhigh | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|
| Arena Visioneffort not stated | – | – | – | – | 1270#44 | Arena (LMArena) | |
| LiveBench Designnot in indexeffort not stated | – | – | – | – | 75.9%#28 | LiveBench | |
| MMMU-Pro | 69.9%#151 | 73.8%#121 | 74.2%#115 | 76.3%#91 | – | Artificial Analysis | |
| MMMU-Pro (Vals)not in index | – | – | – | 83.9%#28 | – | Vals AI | |
| MortgageTaxnot in index | – | – | – | 64.9%#44 | – | Vals AI | |
| SAGEnot in index | – | – | – | 52.4%#7 | – | Vals AI |
Long context
| Benchmark | no reasoning | low | medium | xhigh | Source | Trend |
|---|---|---|---|---|---|---|
| AA-LCR | 69.3%#226 | 77.3%#132 | 79.7%#88 | 82.0%#47 | Artificial Analysis | |
| MLCRnot in index | 10.0%#113 | 7.8%#128 | 3.9%#154 | 22.8%#34 | Artificial Analysis |
Composite
| Benchmark | no reasoning | low | medium | xhigh | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | – | – | 151.7#53 | Epoch AI Benchmarking Hub | |
| LiveBench 2.0effort not stated | – | – | – | – | 62.7%#36 | LiveBench | |
| LiveBench 1.0not in indexeffort not stated | – | – | – | – | 75.3%#38 | LiveBench | |
| AA Intelligence Index v4.3.2 | 20.1#234 | 26.2#156 | 27.6#142 | 33.7#104 | – | Artificial Analysis | |
| AA Cyber Indexnot in index | – | – | – | 12.7#20 | – | 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 |
|---|---|---|---|---|---|---|
| Cerebras | 349 tok/s | 0.27 s | $0.990 | $1.49 | 66k | fp16 |
| ModelRun [by Modular] | 184 tok/s | 0.20 s | $0.200 | $2.50 | 262k | fp4 |
| Modal | 151 tok/s | 1.03 s | $0.040 | $2.30 | 262k | – |
| DekaLLM | 122 tok/s | 0.63 s | $0.040 | $2.50 | 262k | – |
| Phala | 88 tok/s | 1.04 s | $0.200 | $2.50 | 1M | – |
| Parasail | 87 tok/s | 0.46 s | $0.240 | $2.20 | 262k | fp8 |
| NEAR AI | 73 tok/s | 1.46 s | $0.040 | $1.35 | 262k | fp8 |
| Venice | 71 tok/s | 1.85 s | $0.450 | $3.20 | 262k | fp8 |
| AkashML | 58 tok/s | 0.42 s | $0.225 | $1.98 | 262k | fp8 |
| Tenstorrent | 56 tok/s | 0.89 s | $0.020 | $0.500 | 262k | fp4 |
| Reka AI | 54 tok/s | 0.69 s | $0.040 | $2.20 | 262k | fp8 |
| Alibaba Cloud Int. | 47 tok/s | 2.81 s | $0.425 | $2.55 | 1M | – |
| Wafer | 46 tok/s | 2.09 s | $0.060 | $2.30 | 262k | fp8 |
| NovitaAI | 42 tok/s | 5.43 s | $0.420 | $3.00 | 1M | – |
| Ionstream | 41 tok/s | 1.64 s | $0.040 | $1.95 | 262k | fp8 |
| DeepInfra | 36 tok/s | 0.94 s | $0.150 | $1.88 | 262k | bf16 |
| Mancer | 32 tok/s | 0.62 s | $0.200 | $2.50 | 262k | fp8 |
| Chutes | 32 tok/s | 2.27 s | $0.039 | $1.25 | 262k | fp8 |
| CoreWeave | 28 tok/s | 1.57 s | $0.400 | $3.00 | 262k | fp8 |
| Darkbloom | 28 tok/s | 6.89 s | $0.035 | $2.20 | 262k | fp4 |
Price history
The price above, per 1M tokens, every day it changed.
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 at $0.100 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0011 | $0.0010 | 54.7 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0078 | $0.0042 | 1.0 min |
| Code edit | 6,000 / 1,500 | $0.0075 | $0.0057 | 1.4 min |
| Agentic coding session | 60,000 / 4,000 | $0.042 | $0.024 | 2.3 min |
| Structured extraction | 2,000 / 200 | $0.0016 | $0.0010 | 52.5 s |
See also
Questions people ask
- What are Qwen3.8 27B's benchmark scores?
- As of 24 Oct 2026, Qwen3.8 27B (xhigh effort) scores 56.2 ±6.6 on the BenchLeader Index, strongest in composite (67) and weakest in agents & tools (47). That rests on 18 independent benchmark results, and every score on this page links to its publisher.
- How does Qwen3.8 27B rank?
- Qwen3.8 27B (xhigh effort) is #211 of 785 ranked configurations on 24 Oct 2026. 327 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 Qwen3.8 27B cost per 1M tokens?
- As of 24 Oct 2026, Qwen3.8 27B costs $0.500 per 1M input tokens and $3.00 per 1M output tokens, $1.13 blended (three input tokens per output token). The list price last changed on 6 Oct 2026.
- When was Qwen3.8 27B released?
- Alibaba released Qwen3.8 27B on 14 Aug 2026. BenchLeader has results for it at 5 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Qwen3.8 27B: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/qwen3-8-27b.