Alibaba Qwen · Benchmarks

Qwen3.8 27B benchmarks

Qwen3.8 27B has 7 published scores in this dataset, from a Alibaba Qwen release dated 2026-08-14. Its composite index is 83.2, which places it 13th of 22 ranked models and 8th among 15 open-weight models. Its strongest area is knowledge (index 97.3) and its weakest is reasoning (index 70.3). Its best placing is on LiveCodeBench, where its 90.3 ranks 2nd of 7 models with a published score.

Deploy hereDeploy Qwen3.8-27B-Uncensored-FP8Official page

OpenLLM Buddy runs Qwen3.8-27B-Uncensored-FP8, an uncensored fine-tune of Qwen3.8 27B, on H100 80GB (80 GB VRAM). The scores below are the official base model's; a fine-tune can score slightly differently.

Category profile

Index 0–100: each score relative to the best published score on that test.

ReasoningCodingAgenticKnowledge
Reasoning
70.3#12 of 16 · 2 tests
Coding
81.9#7 of 13 · 3 tests
Agentic
—no data
Knowledge
97.31 test · not ranked

Model facts

Lab
Alibaba Qwen
Type
OpenLLM Buddy catalog
Released
2026-08-14
Parameters
27B dense
Context
262K (up to 1M)
License
Apache-2.0
Runs on
H100 80GB

Qwen3.8 27B hardware requirements and cost

At 4-bit it needs about 19 GB, which fits a single RTX 4090, so it can run on a home workstation with llama.cpp, Ollama, or vLLM. Full BF16 precision needs about 62 GB (1× H100 (80 GB)).

Estimated GPU memory for Qwen3.8 27B
PrecisionWeightsWith headroomSmallest setup
BF16 (full precision)54 GB62 GB1× H100 (80 GB)
FP8 / 8-bit27 GB31 GBRTX 5090 (32 GB)
4-bit (GGUF Q4_K_M)16 GB19 GBRTX 4090 (24 GB)

Estimate: parameters × bytes per parameter, plus 15% for the KV cache and runtime at modest context lengths. Long contexts and large batches need more.

On OpenLLM Buddy: Qwen3.8-27B-Uncensored-FP8 runs on H100 80GB as an FP8 build, with an OpenAI-compatible API and no per-token fees. Prices start at $53 USD for 11 hours.

  • 11 hours: $53 USD
  • 24 hours: $116 USD
  • 1 week: $809 USD
  • 1 month: $3236 USD
Deploy Qwen3.8 27B

Lighter or stronger open models

Open models ranked just above and below on the composite index, with their 4-bit memory needs.

Qwen3.8 27B speed and pricing

Measured figures link to their source; estimates are labelled.

API output speed
41 tok/sArtificial Analysis median · Artificial Analysis
Time to first token
0.89 sOpenRouter model page provider stats
API price per 1M tokens
$0.425 in · $2.55 outlab's own endpoint on OpenRouter · OpenRouter endpoints API
Speed on OpenLLM Buddy (H100 80GB, FP8)estimate
52–69 tok/s
OpenLLM Buddy GPU rental
from $4.82 per hourflat rate, no per-token fees
Cost per 1M output tokens, one userestimate
≈ $22.07falls sharply when several requests share the GPU
On your own RTX 4090 (4-bit)estimate
28–39 tok/s

Estimate for one user on one GPU: active weights read per token ÷ memory bandwidth at 50–70% efficiency, plus 3 ms overhead per token. Real speed varies with engine, context length and quantization; serving several users at once raises total throughput.

Compare speed and price across all models

Score by test

Rank is among every model with a published score on that test.

GPQA Diamond89.2%Graduate-level, Google-proof science questions in biology, physics, and chemistry.#13 of 19 · 6.9 behind GPT-6 AstraHumanity's Last Exam30.8%Expert-written closed-ended questions at the frontier of human knowledge.#18 of 23 · 33.6 behind Claude Opus 5.5LiveCodeBench90.3%LiveCodeBench v6#2 of 7 · 3.2 behind DeepSeek V4 ProSWE-bench Pro61.7%Harder, longer software tasks across many languages and repositories.#7 of 9 · 28.2 behind Claude Opus 5.5Terminal-Bench73.0%Terminal Bench 2.1 (Terminus harness)#11 of 13 · 17.6 behind DeepSeek V4.1 FlashIFBench79.5%Out-of-distribution instruction-following constraints.#3 of 3 · 2.2 behind Nemotron 3 UltraLMArena Text Elo1436LMArena text overall, 'Qwen3.8-27B', ±5, as of 2026-10-27#18 of 21 · 84 behind Gemini 4 Argon

Nearby in the composite ranking

Models with the closest composite index.

  1. 1Reflection Beam 1.087.0
  2. 2GPT-6.1 Sol86.5
  3. 3Qwen3.6 27B85.8
  4. 4Qwen3.8 27B83.2
  5. 5GLM 5.382.7
  6. 6DeepSeek V4.1 Flash79.8
  7. 7Gemma 4 31B79.2

Head-to-head comparisons

Side-by-side pages built from the tests both models report.

Sources

ModelTestScoreSettingSource
Qwen3.8 27BGPQA Diamond89.2—Qwen3.8-27B model card (Hugging Face)
Qwen3.8 27BHumanity's Last Exam30.8—Qwen3.8-27B model card (Hugging Face)
Qwen3.8 27BLiveCodeBench90.3LiveCodeBench v6Qwen3.8-27B model card (Hugging Face)
Qwen3.8 27BSWE-bench Pro61.7—Qwen3.8-27B model card (Hugging Face)
Qwen3.8 27BTerminal-Bench73.0Terminal Bench 2.1 (Terminus harness)Qwen3.8-27B model card (Hugging Face)
Qwen3.8 27BIFBench79.5—Qwen3.8-27B model card (Hugging Face)
Qwen3.8 27BLMArena Text Elo1436LMArena text overall, 'Qwen3.8-27B', ±5, as of 2026-10-27Arena (LMArena) text leaderboard

Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works