Qwen3.8-Omni-Flash: pricing, specs and benchmarks

Key specs

Provider: Alibaba Cloud (Qwen)
Family: Qwen3.8-Omni
API identifier: qwen3.8-omni-flash
Availability: Available
Release date: 2026-09-17
License: Proprietary
Context window: 1,000,000 tokens
Max output: 131,072 tokens
API access: Yes
Consumer app: Unknown

Capabilities

Inputs

Text input Yes
Image input Yes
Audio input Yes 113 languages and dialects; multichannel audio supported.
Video input Yes Video up to 2 hours / 2 GB, including audio track understanding.
File input Unknown

Outputs

Text output Yes
Image output No
Audio output No Output modality is text; audio output is provided by qwen3.8-omni-flash-realtime.

Features

Reasoning Optional Hybrid thinking mode, enabled by default; reasoning_effort adjustable.
Tool calling Yes
Web search Yes Requires search_strategy 'agent' in Chat Completions; also available via the Responses API web_search tool.

"Unknown" means we have not verified this capability yet; it does not mean the model lacks it.

Pricing

Estimate your monthly cost →

Dimension	Tier	Price	Since	Verified	Source
Input tokens	base	$0.15 per 1M tokens	2026-10-12	2026-10-25	Alibaba Cloud Model Studio model pricing Singapore (International) list price; single input price listed on the pricing page (no per-modality breakdown).
Cached input tokens	base	$0.016 per 1M tokens	2026-10-12	2026-10-25	Alibaba Cloud Model Studio model pricing Cache-hit input price.
Output tokens	base	$0.47 per 1M tokens	2026-10-12	2026-10-25	Alibaba Cloud Model Studio model pricing
Tool and search requests	base	$10.00 per 1,000 requests	2026-10-12	2026-10-25	Web search with large models - Alibaba Cloud Model Studio Singapore: 'agent' search strategy, or the web_search tool via the Responses API or Anthropic-compatible method; per 1,000 calls, plus retrieved content billed as input tokens. No fee is listed for the default turbo/max strategies

Prices are shown in the currency the provider publishes; we never convert currencies.

Benchmarks

Benchmark	Setting	Score	Reported by	Measured	Verified	Source
CharXiv Reasoning accuracy %, higher is better	reasoning questions (RQ), with code interpreter (CI); effort not stated	91.4 %	Self-reported	—	2026-10-25	Qwen3.8-Omni-Flash: Omni Senses. Agentic Delivery. (Qwen blog, September 2026)
CharXiv Reasoning accuracy %, higher is better	reasoning questions (RQ), without code interpreter (CI); effort not stated	83.5 %	Self-reported	—	2026-10-25	Qwen3.8-Omni-Flash: Omni Senses. Agentic Delivery. (Qwen blog, September 2026)
DeepSWE v1.1 % resolved (pass@1), higher is better	v1.1; effort not stated; best of Claude Code and mini-SWE-agent harnesses, temp 1.0, top_p 0.95, 256K context	57.8 %	Self-reported	—	2026-10-25	Qwen3.8-Omni-Flash: Omni Senses. Agentic Delivery. (Qwen blog, September 2026)
Humanity's Last Exam accuracy %, higher is better	tools not stated; effort not stated; GPT-4o judge	36.5 %	Self-reported	—	2026-10-25	Qwen3.8-Omni-Flash: Omni Senses. Agentic Delivery. (Qwen blog, September 2026)

CharXiv Reasoning caveats:

There is no maintained public leaderboard (the project site says it is under construction), so almost all current numbers are self-reported by providers. Results with tools such as a code interpreter are much higher than without: Alibaba reports 83.7% without and 90.2% with tools for Qwen3.8-27B. Top models now exceed the human baseline, so differences at the top are small and may reflect label noise. Descriptive-question scores are a different metric and must not be recorded here.

DeepSWE v1.1 caveats:

The official leaderboard uses mini-swe-agent for all models and shows whole-number percentages with margins of about ±1 to ±6 points. Providers often report scores from their own scaffolds or effort settings; DeepSeek reports 65.5% to 74.2% for the same model depending on the scaffold. DeepSWE v1 scores are not comparable with v1.1. With only 113 tasks, small differences are within noise.

Humanity's Last Exam caveats:

Results differ a lot by setting: with tools versus without tools (Anthropic reports 45.9% without tools and 57.4% with tools for Claude Haiku 5.5), and the full set versus the text-only subset. Providers use different judge models and harnesses (Scale uses o3-mini; Alibaba states GPT-4o). Derived sets such as HLE-Verified (a third-party revision with 1,811 scored items), HLE-Rolling and HLE-Diamond are different item sets and are not comparable. The questions are public, so contamination is possible, and community reviews have reported flawed items.

One row per result, without rank numbers: each result was measured under its own setting (effort level, harness, tools, test version), as stated by the source, and scores compare only within the same setting. Self-reported results come from the provider and have not been reproduced independently.

Compare Qwen3.8-Omni-Flash

Compare with

Guides

About Qwen3.8-Omni-Flash

qwen3.8-omni-flash accepts text, images, audio and video and produces text output. Alibaba Cloud documents audio input in 113 languages and dialects, multichannel (spatial) audio input, function calling, the web_search built-in tool and implicit context caching. Thinking is enabled by default with adjustable reasoning effort. Real-time audio and video conversation with audio output is offered by the separate qwen3.8-omni-flash-realtime model.

Strengths and weaknesses

Strengths

Text, image, audio and video input, including sound within videos
1,000,000-token context window with up to 131,072 output tokens
Hybrid thinking with adjustable reasoning effort, function calling and web search
List price of USD 0.15 input / USD 0.47 output per 1M tokens

Weaknesses

Text output only in this model (audio output requires the separate realtime model)
Built-in tools limited to web_search; structured output limited to JSON Object mode
Not among the models supported for batch inference in the Singapore region