Patchwire

Qwen · Released Aug 14, 2026

Qwen3.8 27B

Qwen3.8 27B is an open-weight dense vision-language model from Qwen. It is suited for coding, professional workflows, research, multimodal interaction, and long-running agent tasks, with flexible thinking that can be...

ReasoningOpen weightsVision
  • Top 24%WebDev Arena
Compare in showdown → OpenRouter ↗ All Qwen models
Released
Aug 14 2026
Context window
1M 66K out
Input / 1M tokens
$0.04 from $0.02
Output / 1M tokens
$1.35 from $0.50

Where to run it

21 providers compared

Cheapest: Tenstorrent at $0.02 / $0.50 per 1M input / output tokens — 88% less than the priciest option.

ProviderInputOutputCache readContextMax outputUptimevs. best
TenstorrentFP4 $0.02 $0.50 $0.009262K66K94.3%24h Cheapest
ChutesFP8 $0.039 $1.25 $0.004262K66K99.9%24h +144%
Near AIFP8 $0.04 $1.35 $0.018262K66K99.9%24h +163%
IonstreamFP8 $0.04 $1.95 $0.02262K66K99.6%24h +270%
DarkbloomFP4 $0.035 $2.20 $0.018262K33K91.3%24h +312%
RekaFP8 $0.04 $2.20 $0.02262K236K100%24h +314%
DeepInfraBF16 $0.15 $1.88 $0.038262K236K100%24h +315%
Modal $0.04 $2.30 $0.02262K236K98.9%24h +332%
WaferFP8 $0.06 $2.30 $0.059262K236K100%24h +343%
DekaLLM $0.04 $2.50 $0.02262K236K99.7%24h +368%
AkashMLFP8 $0.23 $1.98 $0.05262K131K100%24h +374%
ParasailFP8 $0.24 $2.20 $0.05262K236K98.1%24h +421%
ModelRunFP4 $0.20 $2.50 $0.09262K236K99.9%24h +454%
Mancer 2FP8 $0.20 $2.50 —262K236K99.6%24h +454%
Phala $0.20 $2.50 $0.051M262K99.6%24h +454%
Alibaba $0.42 $2.55 $0.0851M131K99.2%24h +583%
CoreWeaveFP8 $0.40 $3.00 $0.15262K236K33.7%24h +650%
Novita $0.42 $3.00 $0.0851M131K99.7%24h +661%
CerebrasFP16 $0.99 $1.49 $0.9966K33K99.8%24h +696%
Cloudflare $0.45 $3.20 $0.05262K236K100%24h +713%
VeniceFP8 $0.45 $3.20 —262K66K100%24h +713%

Value

Price vs. performance

On the value frontierNo cheaper model beats Qwen3.8 27B on SciCode.
27.8%40.8%53.9%66.9% $0.01$0.10$1.00$10.00 DeepSeek V4.1 Flash 1017 (DeepSeek) — 51.8% · $0.30/1M in Gemini 4 Argon (Google) — 61.8% · $2.00/1M in Step 5 Preview (StepFun) — 58.9% · $1.00/1M in Claude Haiku 5.5 (Anthropic) — 55.0% · $0.10/1M in Mistral Large 4 (Mistral) — 54.2% · $0.68/1M in GPT-6.1 Sol (OpenAI) — 55.8% · $2.00/1M in Claude Sonnet 5.5 (Anthropic) — 61.0% · $2.00/1M in GPT-6 Luna (OpenAI) — 54.6% · $0.10/1M in GPT-6 Sol (OpenAI) — 57.6% · $2.00/1M in Claude Opus 5.5 (Anthropic) — 66.9% · $4.00/1M in MiMo-V2.6-Flash (Xiaomi) — 51.3% · $0.14/1M in MiMo-V2.6-Pro (Xiaomi) — 60.9% · $0.43/1M in Grok 4.7 (xAI) — 57.8% · $2.00/1M in DeepSeek V4.1 Flash (DeepSeek) — 51.8% · $0.056/1M in GPT-6 Astra (OpenAI) — 56.5% · $10.00/1M in Muse Spark 1.3 (Meta) — 59.7% · $1.25/1M in Gemini 3.8 Flash (Google) — 56.6% · $0.82/1M in Claude Fable 5.1 (Anthropic) — 63.1% · $10.00/1M in GLM 5.3 Flash (Z.ai) — 51.6% · $0.047/1M in GLM 5.3 (Z.ai) — 59.0% · $0.079/1M in Gemini 3.7 Flash (Google) — 59.8% · $0.75/1M in DeepSeek V4 Pro 0813 (DeepSeek) — 51.0% · $0.66/1M in Grok 4.6 (xAI) — 56.5% · $2.00/1M in Muse Spark 1.2 (Meta) — 57.4% · $1.25/1M in DeepSeek V4 Flash 0731 (DeepSeek) — 49.9% · $0.44/1M in Claude Opus 5 (Anthropic) — 56.4% · $5.00/1M in Gemini 3.6 Flash (Google) — 53.4% · $0.75/1M in Gemini 3.5 Flash Lite (Google) — 41.3% · $0.30/1M in Kimi K3 (Moonshot AI) — 59.5% · $0.24/1M in Muse Spark 1.1 (Meta) — 58.8% · $1.25/1M in GPT-5.6 Luna (OpenAI) — 53.6% · $0.20/1M in GPT-5.6 Terra (OpenAI) — 55.0% · $2.00/1M in GPT-5.6 Sol (OpenAI) — 57.8% · $4.00/1M in Grok 4.5 (xAI) — 55.0% · $2.00/1M in Claude Sonnet 5 (Anthropic) — 54.3% · $2.00/1M in GLM 5.2 (Z.ai) — 51.2% · $0.62/1M in Kimi K2.7 Code (Moonshot AI) — 47.8% · $0.67/1M in Claude Fable 5 (Anthropic) — 61.0% · $10.00/1M in MiniMax M3 (MiniMax) — 47.1% · $0.30/1M in Step 3.7 Flash (StepFun) — 40.1% · $0.20/1M in Claude Opus 4.8 (Anthropic) — 53.5% · $5.00/1M in Grok Build 0.1 (xAI) — 50.2% · $1.00/1M in Gemini 3.5 Flash (Google) — 53.9% · $1.50/1M in Gemini 3.1 Flash Lite (Google) — 43.4% · $0.25/1M in Grok 4.3 (xAI) — 48.3% · $1.25/1M in GPT-5.5 (OpenAI) — 56.1% · $5.00/1M in DeepSeek V4 Pro 0423 (DeepSeek) — 51.0% · $0.95/1M in DeepSeek V4 Flash 0423 (DeepSeek) — 49.9% · $0.091/1M in MiMo-V2.5-Pro (Xiaomi) — 50.2% · $0.43/1M in MiMo-V2.5 (Xiaomi) — 43.1% · $0.14/1M in Kimi K2.6 (Moonshot AI) — 53.5% · $0.65/1M in Claude Opus 4.7 (Anthropic) — 54.5% · $5.00/1M in GLM 5.1 (Z.ai) — 44.8% · $0.97/1M in Gemma 4 31B (Google) — 45.5% · $0.20/1M in MiniMax M2.7 (MiniMax) — 50.1% · $0.30/1M in GPT-5.4 Nano (OpenAI) — 47.2% · $0.20/1M in GPT-5.4 Mini (OpenAI) — 52.1% · $0.75/1M in Mistral Small 4 (Mistral) — 27.8% · $0.15/1M in GPT-5.4 (OpenAI) — 56.6% · $2.50/1M in Gemini 3.1 Flash Lite Preview (Google) — 43.4% · $0.25/1M in Gemini 3.1 Pro Preview (Google) — 58.7% · $2.00/1M in Claude Sonnet 4.6 (Anthropic) — 46.8% · $3.00/1M in Kimi K2.5 (Moonshot AI) — 49.0% · $0.50/1M in GLM 4.7 (Z.ai) — 45.1% · $0.54/1M in Mistral Large 3 2512 (Mistral) — 36.6% · $0.50/1M in DeepSeek V3.2 (DeepSeek) — 38.9% · $0.28/1M in GPT-5.1 (OpenAI) — 43.3% · $1.25/1M in Claude Haiku 4.5 (Anthropic) — 42.3% · $1.00/1M in GLM 4.6 (Z.ai) — 38.4% · $0.55/1M in Claude Sonnet 4.5 (Anthropic) — 45.7% · $3.00/1M in DeepSeek V3.2 Exp (DeepSeek) — 38.9% · $0.27/1M in DeepSeek V3.1 Terminus (DeepSeek) — 38.0% · $0.27/1M in GPT-5 (OpenAI) — 42.9% · $1.25/1M in GPT-5 Mini (OpenAI) — 39.2% · $0.25/1M in gpt-oss-120b (OpenAI) — 36.0% · $0.15/1M in gpt-oss-20b (OpenAI) — 38.9% · $0.018/1M in Gemini 2.5 Pro (Google) — 46.3% · $1.25/1M in Gemini 2.5 Pro Preview 06-05 (Google) — 46.3% · $1.25/1M in R1 0528 (DeepSeek) — 38.3% · $0.50/1M in Qwen3.8 Max (0902) (Qwen) — 53.2% · $2.00/1M in Qwen3.7 Plus (Qwen) — 46.1% · $0.32/1M in Qwen3.7 Max (Qwen) — 49.5% · $1.48/1M in Qwen3.6 35B A3B (Qwen) — 36.6% · $0.10/1M in Qwen3.6 27B (Qwen) — 37.3% · $0.25/1M in Qwen3.6 Plus (Qwen) — 40.7% · $0.33/1M in Qwen3.5-9B (Qwen) — 29.5% · $0.10/1M in Qwen3.5-35B-A3B (Qwen) — 29.3% · $0.15/1M in Qwen3.5-122B-A10B (Qwen) — 35.6% · $0.26/1M in Qwen3 Coder Next (Qwen) — 36.2% · $0.12/1M in Qwen3 235B A22B Thinking 2507 (Qwen) — 41.4% · $0.45/1M in Qwen3 14B (Qwen) — 31.6% · $0.12/1M in Qwen3 32B (Qwen) — 35.4% · $0.08/1M in Qwen3.8 27B (Qwen) — 46.6% · $0.04/1M in Qwen3.8 27B

Qwen3.8 27B Other Qwen Other labs Value frontier · SciCode vs. input price per 1M tokens (log scale) · Drag to pan, pinch or Ctrl + scroll to zoom, double-click to zoom in.

Benchmarks

8 benchmark scores

BenchmarkScoreRankLeader
SciCodeSoftware engineering · Research-coding problems from the natural sciences 46.6% #58of 93 Claude Opus 5.566.9%
APEX-AgentsAgents & computer use · Long professional tasks in banking, consulting and law 47.5% #29of 51 Gemini 4 Argon82.4%
WebDev ArenaDesign & visual · People vote on which model builds the better website (Elo) 1593 #21of 88 Claude Opus 5.51820
ARC-AGI-2Reasoning & knowledge · Few-shot abstract reasoning on grid puzzles 42.4% #41of 63 GPT-6 Astra95.0%
ARC-AGI-1Reasoning & knowledge · Pattern induction from input–output examples 87.5% #36of 63 Gemini 4 Argon98.5%
CritPtReasoning & knowledge · Open-ended, research-level physics problems 5.4% #57of 96 GPT-5.6 Sol32.3%
LMCAReasoning & knowledge · Judge the quality of conceptual arguments like experts 41.4% #49of 96 Claude Sonnet 5.568.3%
ProofBenchMaths · Graduate maths proofs verified in Lean 4 16.0% #56of 70 Claude Sonnet 5.5100.0%

Specs

Details

Model ID
qwen/qwen3.8-27b
Developer
Qwen
Release date
Aug 14, 2026 ·
Context window
1,000,000 tokens
Max output
65,536 tokens
API price
$0.04 / $1.35 per 1M input / output tokens
Providers
21
Input types
text, image, video
Reasoning
Yes
Tool calling
Yes
Open weights
Yes
Free tier
No

Release date, prices and providers from OpenRouter's catalog; buying directly from a provider can cost a little differently. Uptime is the share of OpenRouter requests each provider served over the last day. Benchmark scores are the best result per benchmark from Epoch AI, licensed CC BY 4.0. Rank counts every model tracked here with a score.