Data as of November 12, 2026 · How the score is built
Qwen3.8-27B
Decision readingQwen3.8-27B scores 54.1 out of 100 and ranks #72 of 215. This profile shows 52 source-displayable benchmark rows; its strongest eligible category is Agentic at #24. Published weights can be self-hosted, but infrastructure cost varies and is not a comparable API token rate.
Email me when its price, rank or availability changes.
Capability
54.1/100
field median 49.4#72 of 215 ranked models
Public
#72of 215
Verified #42 of 81
Price
Self-hosted; infrastructure cost varies
input median $0.60No comparable first-party hosted token rate
Speed
40tok/s
field median 92 tok/sFirst token 54.01 s
Context
262Ktokens
field median 256,000Maximum output length is tracked separately
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Strongest published evidence
Agentic ranks #24. Particularly useful for coding agents, browser research, and computer-use workflows.
Validate before choosing
52 published rows leave some tracked benchmark slots empty. No comparable first-party API token rate is published.
Source-linked · 52 displayable benchmark rows
Is Qwen3.8-27B right for your job?
Its place on the LLM Selector’s job rankings, the same order the job pages publish. Describe your own job to see where it stands for that.
Where Qwen3.8-27B ranks by benchmark score:
We keep what you type here, drafts included, for 180 days, and after that only the job and requirements read from it. Nothing is tied to your IP address or account. Leave out names and other personal details. Privacy
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticRank #24 of 135Percentile 83rdWeight 22%9 benchmarksVerified | 61.8 | 9 benchmarks | Verified | |||
| CodingRank #58 of 162Percentile 65thWeight 20%9 benchmarksVerified | 45.9 | 9 benchmarks | Verified | |||
| ReasoningRank #70 of 160Percentile 57thWeight 17%Not listedUnlisted sources | 43.2 | Not listed | Unlisted sources | |||
| MultimodalRank #33 of 67Percentile 52ndWeight 12%12 benchmarksVerified | 60.5 | 12 benchmarks | Verified | |||
| KnowledgeRank #79 of 180Percentile 56thWeight 12%6 benchmarksVerified | 50.1 | 6 benchmarks | Verified | |||
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| Inst. FollowingRank #44 of 157Percentile 72ndWeight 5%1 benchmarkVerified | 86.9 | 1 benchmark | Verified | |||
| MathWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured |
52 of 906 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic9/9 verified
- Coding9/9 verified
- ReasoningNot measured
- Multimodal12/12 verified
- Knowledge6/6 verified
- MultilingualNot measured
- Inst. Following1/1 verified
- MathNot measured
Capability shape
Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.
Qwen3.8-27B category percentile values
- Agentic83rd percentile
- Coding65th percentile
- Reasoning57th percentile
- Multimodal52nd percentile
- Knowledge56th percentile
- MultilingualNot eligible
- Instruction following72nd percentile
- MathNot eligible
The dashed outline is median of 6 nearest peers.
Eligible category ranks
- Agentic#24/135
- Coding#58/162
- Reasoning#70/160
- Multimodal#33/67
- Knowledge#79/180
- MultilingualNot ranked
- Inst. Following#44/157
- MathNot ranked
Benchmark ledger
Every result says who ran the test. For this model, Alibaba ran 27 of 52. Coding opens by default. The tick marks the best source-verified result in the catalog; provisional leaders do not set the reference.
Coding9 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| SWE-bench Pro | 61.7% | Best verified: Claude Fable 5.2 · 93% | Gap31.3 behind | Reported by Alibaba | Counts26% ref. weight |
| DeepSWE | 42.2% | Best verified: Muse Wonder Full · 79.0% | Gap36.8 behind | Reported by Alibaba | Counts15% ref. weight |
| LiveCodeBench (Vals)LiveCodeBench, Vals AI run | 84.0% | Best verified: Claude Fable 5.1 · 90.5% | Gap6.5 behind | Independent run: Vals AI | Counts8% ref. weight |
| Terminal-Bench 2.1 | 73.0% | Best verified: SWE-2 · 92.8% | Gap19.8 behind | Reported by Alibaba | CountsScored in Agentic |
| NL2Repo | 42.3% | Best verified: DeepSeek V4.1 Flash 1017 · 68.2% | Gap25.9 behind | Reported by Alibaba | Shown only |
| LiveCodeBench v6 | 90.3% | Best verified: Sakana Fugu-Ultra · 93.2% | Gap2.9 behind | Reported by Alibaba | Shown only |
| VulcanBench v3 | 82.6% | Best verified: Grok 4.5 · 89.9% | Gap7.3 behind | Run by VulcanBench v3 | Shown only |
| SWE-bench (Vals)SWE-bench, Vals AI run | 86.0% | Best verified: Claude Opus 5 · 97.0% | Gap11 behind | Independent run: Vals AI | Shown only |
| Bug Hunt Bench | 15.0 fixes | Best verified: Claude Fable 5.2 · 58.0 fixes | Gap43 behind | Run by Bug Hunt Bench | Shown only |
Agentic9 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| Terminal-Bench 2.1 | 73.0% | Best verified: SWE-2 · 92.8% | Gap19.8 behind | Reported by Alibaba | Counts9% ref. weight |
| OSWorld-Verified | 84.3% | Best verified: Qwen3.8 Max · 86.1% | Gap1.8 behind | Reported by Alibaba | Counts8% ref. weight |
| JobBench | 33.4% | Best verified: Muse Wonder Full · 75.3% | Gap41.9 behind | Reported by Alibaba | Counts6% ref. weight |
| Terminal-Bench 2.1 (Vals)Terminal-Bench 2.1, Vals AI run | 58.4% | Best verified: Claude Opus 5.5 · 87.6% | Gap29.2 behind | Independent run: Vals AI | Counts4% ref. weight |
| Agents' Last Exam | 42.9% | Best verified: GPT-6.1 Terra · 60.0% | Gap17.1 behind | Reported by Alibaba | Counts3% ref. weight |
| CoWorkBench | 70.7% | Best verified: Qwen3.8-Omni-Flash · 75.3% | Gap4.6 behind | Reported by Alibaba | Shown only |
| WebArena-VerifiedWebArena-Verified Browser Agent Benchmark | 64.8% | Best verified: Muse Spark 1.1 · 69% | Gap4.2 behind | Reported by Alibaba | Shown only |
| AndroidWorld | 81.9% | Best verified: Qwen3.8-Omni-Flash · 87.1% | Gap5.2 behind | Reported by Alibaba | Shown only |
| MazeBenchMazeBench: visual spatial reasoning in a 3D open world | 0% | Best verified: GPT-6 Astra · 23% | Gap23 behind | Run by MazeBench | Shown only |
Multimodal12 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| CharXivCharXiv Reasoning | 90.2% | Best verified: Qwen3.8 Max · 93.5% | Gap3.3 behind | Reported by Alibaba | Counts17% ref. weight |
| ParseBenchParseBench: document parsing for AI agents (LlamaIndex leaderboard) | 69.03% | Best verified: Claude Opus 5.5 · 78.80% | Gap9.8 behind | Run by ParseBench | Counts7% ref. weight |
| CharXiv w/o toolsCharXiv Reasoning without tools | 83.7% | Best verified: Kimi K3.1 Pro · 89% | Gap5.3 behind | Reported by Alibaba | Counts7% ref. weight |
| ExtractBench (LlamaIndex)ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard) | 89.75% | Best verified: GPT-6.1 Sol · 95.31% | Gap5.6 behind | Run by ExtractBench (LlamaIndex) | Shown only |
| MathVision | 90.0% | Best verified: Qwen3.8 Max · 95.2% | Gap5.2 behind | Reported by Alibaba | Shown only |
| MathVision w/ PythonMathVision with Python | 94.6% | Best verified: Kimi K3 · 97.8% | Gap3.2 behind | Reported by Alibaba | Shown only |
| BabyVision | 65.7% | Best verified: Qwen3.8 Max · 82.0% | Gap16.3 behind | Reported by Alibaba | Shown only |
| BabyVision w/ PythonBabyVision with Python | 85.6% | Best verified: Kimi K3.1 Pro · 95.4% | Gap9.8 behind | Reported by Alibaba | Shown only |
| Vision2Web | 62.9% | Best verified: LongCat-2.5 · 78.6% | Gap15.7 behind | Reported by Alibaba | Shown only |
| OmniDocBench 1.5 | 91.1% | Best verified: Qwen3.8 Max · 92.1% | Gap1 behind | Reported by Alibaba | Shown only |
| RealWorldQA | 85.9% | Best verified: Qwen3.8-Flash-Next · 88.5% | Gap2.6 behind | Reported by Alibaba | Shown only |
| ERQA | 65.5% | Best verified: Qwen3.8 Max · 77.8% | Gap12.3 behind | Reported by Alibaba | Shown only |
Knowledge6 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| HLEHumanity's Last Exam | 30.8% | Best verified: Claude Fable 5.2 · 74.2% | Gap43.4 behind | Reported by Alibaba | Counts44% ref. weight |
| HLE w/o toolsHumanity's Last Exam without tools | 30.8% | Best verified: Claude Fable 5.2 · 73.2% | Gap42.4 behind | Reported by Alibaba | Counts7% ref. weight |
| MMLU-Pro (Vals)MMLU-Pro, Vals AI run | 84.3% | Best verified: Claude Fable 5.1 · 92.4% | Gap8.1 behind | Independent run: Vals AI | Counts6% ref. weight |
| GPQAGraduate-Level Google-Proof Q&A | 89.2% | Best verified: GPT-6 Astra · 96% | Gap6.8 behind | Reported by Alibaba | Counts3% ref. weight |
| GPQA Diamond (Vals)GPQA Diamond, Vals AI run | 88.9% | Best verified: Gemini 3.1 Pro · 95.5% | Gap6.6 behind | Independent run: Vals AI | Counts2% ref. weight |
| GPQA-DGPQA Diamond | 89.2% | Best verified: GPT-6 Astra · 96.0% | Gap6.8 behind | Reported by Alibaba | Shown only |
Inst. Following1 row
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| IFBenchInstruction Following Benchmark | 79.5% | Best verified: Flint · 89.2% | Gap9.7 behind | Reported by Alibaba | CountsWeighted 70% |
External signals15 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| NYT ConnectionsNYT Connections LLM Benchmark (extended version) | 54.5% | Best verified: Gemini 4 Argon · 98.6% | Gap44.1 behind | Run by NYT Connections | Shown only |
| StealthMarkStealthMark Intelligence Benchmarks (overall score) | 51.5 | Best verified: GPT-6 Astra · 74.6 | Gap23.1 behind | Run by StealthMark | Shown only |
| BabaIsBenchBabaIsBench (Baba Is You intro levels, pass@1) | 21% | Best verified: Claude Opus 5 · 100% | Gap79 behind | Run by BabaIsBench | Shown only |
| WinoGrande (Perplexity panel) | 73.10% | Best verified: Jev 1.13.0 · 90.70% | Gap17.6 behind | Lab-reported, from another publisher’s document | Shown only |
| FinancialPhraseBank (Perplexity panel) | 75.68% | Best verified: Perplexity Decider v1 27B · 84.18% | Gap8.5 behind | Lab-reported, from another publisher’s document | Shown only |
| RAGTruth (Perplexity panel) | 61.53% | Best verified: Perplexity Decider v1 27B · 88.80% | Gap27.3 behind | Lab-reported, from another publisher’s document | Shown only |
| JudgeBench (Perplexity panel) | 68.86% | Best verified: Jev 1.13.0 · 78.57% | Gap9.7 behind | Lab-reported, from another publisher’s document | Shown only |
| BBH (Perplexity panel) | 72.80% | Best verified: Jev 1.13.0 · 94.27% | Gap21.5 behind | Lab-reported, from another publisher’s document | Shown only |
| JevBench public hard (Perplexity panel) | 72.28% | Best verified: Jev 1.13.0 · 73.27% | Gap1 behind | Lab-reported, from another publisher’s document | Shown only |
| TabFact (Perplexity panel) | 78.60% | Best verified: Perplexity Decider v1 27B · 90.60% | Gap12 behind | Lab-reported, from another publisher’s document | Shown only |
| ContractNLI (Perplexity panel) | 80.78% | Best verified: Qwen3.8-27B · 80.78% | GapHighest listed | Lab-reported, from another publisher’s document | Shown only |
| Circa (Perplexity panel) | 87.00% | Best verified: Perplexity Decider v1 27B · 89.20% | Gap2.2 behind | Lab-reported, from another publisher’s document | Shown only |
| Belebele (Perplexity panel) | 93.20% | Best verified: Jev 1.13.0 · 95.00% | Gap1.8 behind | Lab-reported, from another publisher’s document | Shown only |
| TruthfulQA binary (Perplexity panel) | 82.80% | Best verified: Jev 1.13.0 · 92.00% | Gap9.2 behind | Lab-reported, from another publisher’s document | Shown only |
| Perplexity Decision Panel | 74.76% | Best verified: Perplexity Decider v1 27B · 85.71% | Gap10.9 behind | Lab-reported, from another publisher’s document | Shown only |
Outlined bars are lab-reported. Filled bars are benchmark-run or independent.
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Base entry
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- Not publishedQwen3.8-27B model card
- Context window
- 262KQwen3.8-27B model card
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- text, image, videoQwen3.8-27B model card
- Output modalities
- textQwen3.8-27B model card
- Parameters
- Not sourced yet
- Availability
- Qwen publishes the 27B dense checkpoint and a separate FP8 artifact on Hugging Face under Apache-2.0. LM Studio lists a community GGUF package for local use. Qwen Cloud says a hosted qwen3.8-27b service with a default 1M context window and built-in tools is coming soon, so no current first-party API identifier or token rate is stored.
- Cloud regions
- Not tracked yet
- Lifecycle
- Current
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Not documented in the pricing recordQwen3.8-27B model card
- Self-host
- Qwen3.8-27B needs ~24GB VRAM (1× NVIDIA RTX 4090 (24GB)).
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
Qwen3.8-27B ranks #72 of 215 on the public leaderboard with a score of 54.06/100. Its source-verified position is #42 of 81.
Qwen3.8-27B is a open weight model with a 262K context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Qwen publishes the 27B dense checkpoint and a separate FP8 artifact on Hugging Face under Apache-2.0. LM Studio lists a community GGUF package for local use. Qwen Cloud says a hosted qwen3.8-27b service with a default 1M context window and built-in tools is coming soon, so no current first-party API identifier or token rate is stored.
Official exact-value snapshot from Qwen's Qwen3.8-27B model card. We keep Terminal-Bench 2.1 and LiveCodeBench v6 separate from the older weighted lanes, preserve no-Code-Interpreter and Code-Interpreter visual results on separate keys, and leave QwenSWEBench, RecreationBench, ClawEval-MM, and SWE-MM outside the schema instead of forcing them into non-equivalent fields. Qwen reports a dense 27B native vision-language model with a 262,144-token native context window extensible to 1,000,000 tokens.
Its explicit predecessor is Qwen3.6-27B. 52 of 906 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Agentic at #24, while its lowest eligible position is Knowledge at #79. particularly useful for coding agents, browser research, and computer-use workflows.
Last updated November 12, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Where to run Qwen3.8-27B by API
6 hosts publish a per-token price for it. Cerebras has the lowest output rate, $1.49 per million tokens. Prices read from each host’s own pages on November 11, 2026.
| Host | Input /1M | Output /1M | Cached /1M | Context | Listed speed |
|---|---|---|---|---|---|
| Cerebras | $0.99 | $1.49 | — | 128K | 1,850 tok/s |
| DeepInfra | $0.15 | $1.88 | $0.037 | 262K | — |
| OpenRouter | $0.15 | $1.88 | $0.037 | 1.05M | — |
| Novita AI | $0.42 | $3.00 | $0.085 | 1M | — |
| Nebius Token Factory | $0.45 | $3.00 | $0.10 | 1M | 218 tok/s |
| Groq | $0.80 | $4.00 | — | 131K | 450 tok/s |
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Questions
How does Qwen3.8-27B perform overall in AI benchmarks?
Qwen3.8-27B ranks #72 out of 215 models on the public BenchAlign leaderboard, with a score of 54.06/100. Its evidence status is Supported, and this profile shows 52 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.
Is Qwen3.8-27B good for knowledge and understanding?
Qwen3.8-27B ranks #79 out of 180 eligible models for knowledge and understanding, with a public category score of 50.1/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Qwen3.8-27B good for coding and programming?
Qwen3.8-27B ranks #58 out of 162 eligible models for coding and programming, with a public category score of 45.9/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Qwen3.8-27B good for agentic tool use and computer tasks?
Qwen3.8-27B ranks #24 out of 135 eligible models for agentic tool use and computer tasks, with a public category score of 61.8/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Qwen3.8-27B good for multimodal and grounded tasks?
Qwen3.8-27B ranks #33 out of 67 eligible models for multimodal and grounded tasks, with a public category score of 60.5/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Qwen3.8-27B good for instruction following?
Qwen3.8-27B ranks #44 out of 157 eligible models for instruction following, with a public category score of 86.9/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Qwen3.8-27B open source?
Qwen3.8-27B is an open-weight model from Alibaba. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Does Qwen3.8-27B have full benchmark coverage on BenchLM?
No. Qwen3.8-27B currently has 78 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of Qwen3.8-27B?
Qwen3.8-27B has a documented context window of 262K. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.