Models · Qwen (Alibaba) · Out sinceReleased 26 Aug 2026

Qwen3.8 Flash

Available on OpenRouterOn OpenRouterAvailable nowGenerally availableCan run on your own serversOpen weights

Qwen3.8 Flash is made by Qwen (Alibaba). We don't have enough test results yet to rank it. It's cheap to use. Its makers have published it, so you can run it on your own servers.

Production (1M ctx, built-in tools) version of open-weight Qwen3.8-Flash-Next: 125B main model + 51B n-gram embeddings, 6B active; preview of the Qwen4 architecture. Blog header shows 2026-08-03 but HF repo was created 2026-08-24 and OpenRouter listed it 2026-08-26; release date set to 2026-08-26. Benchmarks in the blog are for Qwen3.8-Flash-Next.

Writing & creativity: —
Research & analysis: —
Coding: —
Price: Cheap$0.15 / $0.47
Price per 1M (blended)Blended / 1M: $0.23
MemoryContext: 1M
Longest answerMax output: 131K
Test resultsResults: 9 (0 independent0 indep.)
Out sinceReleased: 26 Aug 2026
Made byVendor: Qwen (Alibaba)
UnderstandsInputs: text, image, video

Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.

Thinking levels

Reasoning settings

No reasoningDefault

Links

Using it through OpenRouter

On OpenRouter

OpenRouter is a service that gives access to many AI models in one place. This model is listed there as qwen/qwen3.8-flash.

Route it as qwen/qwen3.8-flash at $0.15 in / $0.47 out per 1M tokens, 1M context. Listed since 26 Aug 2026.

frequency_penaltyinclude_reasoninglogprobsmax_tokenspresence_penaltyreasoningresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p

Run it yourself

Self-hosting

On your own servers,Licence, sizeunder your own control.and hardware.

Compare all open models →All open-weights models →

Details comingselfHost data pending

This model can be downloaded and run on your own servers. We're collecting its licence, size and hardware needs now.Open-weights model; licence, parameter count, hardware tier and base-model availability will appear after the next data run.

Every test result

Every result

All the numbers,with where they came from.

Every result we've found for this model, with who measured it and a link to where we read it.

All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.

All results for this model
						NotesNotes
Agents' Last Exam	24.3%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	—	Pass@1; score 51.2; Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
DeepSWE	58.7%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	mini-SWE-agent	DeepSWE 1.1; best of Claude Code and mini-SWE-agent (mini-SWE best); Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
GPQA Diamond	91.7%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	—	Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
Humanity's Last Exam	35.9%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	—	judged by GPT-4o; Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
LiveCodeBench	91.9%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	—	LiveCodeBench v6; Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
NL2Repo-Bench	48.1%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	Claude Code	Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
SWE-bench Multilingual	81.0%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	mini-SWE-agent	Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
SWE-Bench Pro (public, v1)	62.5%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	Claude Code	Qwen-corrected SWE-bench Pro, 256K ctx; Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)
Toolathlon-Verified	73.5%	Defaultthinking (setting not stated)	Maker's own figureVendor-reportedQwen (Alibaba) ↗	26 Aug 2026	—	pass@1; Measured on open-weight Qwen3.8-Flash-Next (the production Qwen3.8-Flash adds 1M ctx and built-in tools)