Qwen3 Omni 30B A3B
Qwen3 Omni 30B A3B is an omni-modal chat model from Alibaba's Qwen3 family, using a mixture-of-experts design with roughly 30B total and about 3B active parameters per token.
API Pricing
| Provider | Input / 1M | Output / 1M |
|---|---|---|
| $0.400 | $0.800 |
Prices updated daily. Last check: Oct 14, 2026
Compare API pricing for every Qwen model →Qwen3 Omni 30B A3B pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Reasoning & Knowledge
- MMLU-Pro72.5%
- GPQA Diamond62.0%
- Humanity's Last Exam4.6%
Coding
- LiveCodeBench42.2%
Math
- AIME 202552.3%
Agentic & Tool Use
- Terminal-Bench Hard1.5%
- τ²-bench16.4%
Instruction & Long Context
- IFBench31.2%
- Long-Context Reasoning0.0%
Benchmarks measured Oct 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- Alibaba
- Modalities
- Text
Capabilities
- Open Source
- Yes
Strengths & Limitations
Strengths
- Mixture-of-experts design with roughly 3B active parameters per token from a ~30B total pool, keeping per-token compute low relative to total capacity
- Measured output throughput of about 104.8 tokens per second, suitable for streaming chat interfaces
- Time to first token of roughly 916 ms in benchmark testing, keeping perceived response latency under a second
- Part of Alibaba's Qwen3 family, so it shares tokenizer and prompt-format conventions with other Qwen3 variants for easier swapping
- "Omni" variant positioning targets mixed-modality conversational input rather than text-only workloads
- Mid-size footprint makes it a candidate for self-hosting where larger flagship models are impractical
Limitations
- We do not track a confirmed context window for this model — verify the limit with your chosen provider
- Modality support for omni-modal builds varies by deployment; not every host enables every input track
- No verified accuracy benchmark scores in our data, so quality comparisons against peers require independent testing
- Only about 3B parameters are active per token, which can limit performance on the hardest reasoning tasks relative to larger-activation models
- 30B total parameters still require substantial GPU memory when self-hosted, despite the low active count
- Provider availability for omni-modal Qwen variants is typically narrower than for mainstream text models
Key Features
About Qwen3 Omni 30B A3B
Common Use Cases
Qwen3 Omni 30B A3B fits interactive assistant workloads where mixed-modality input needs to be handled by a single endpoint at conversational speed — voice or media-aware chat front ends, multimodal customer support triage, and content description or summarization pipelines that accept more than plain text. Its measured sub-second time to first token and triple-digit output rate make it reasonable for streaming UIs where users watch text appear. The sparse mixture-of-experts design also suits high-volume request patterns, since per-token compute is closer to a small model than its 30B total parameter count suggests, which matters both for hosted throughput and for self-hosted batching. For tasks demanding maximum reasoning depth — competition math, long multi-step agentic planning, or large-repository code work — a larger-activation or dedicated reasoning model is the more conservative choice, and buyers should benchmark this model on their own task set given that we do not track verified accuracy scores for it.
Frequently Asked Questions
What does "30B A3B" mean in the model name?
It follows Qwen's naming convention for mixture-of-experts checkpoints: roughly 30 billion total parameters, with approximately 3 billion active per token. Only a subset of experts fires on each token, so inference compute per token is far lower than the total parameter count implies, while memory still has to hold the full weight set when self-hosting.
How much does Qwen3 Omni 30B A3B cost to use?
Pricing varies by provider and by pricing type — hosted per-token API rates differ from serverless and dedicated-capacity arrangements, and input and output tokens are usually billed at different rates. Check the pricing table on this page for current per-provider figures rather than relying on any fixed number.
What is this model best used for?
Interactive multimodal chat and assistant workloads that need fast first-token response and steady streaming output. Its measured ~916 ms time to first token and ~104.8 tokens per second output rate support real-time interfaces, and the sparse architecture helps with high request volumes.
Which input modalities does it support?
The "Omni" designation indicates a multimodal variant rather than a text-only one, but we do not track a confirmed per-modality list for this entry, and hosted deployments of omni-modal models often enable only some input tracks. Confirm supported inputs directly with the provider you plan to use.
How does it compare to other Qwen3 variants?
It sits in the mid-size, sparsely activated part of the Qwen3 lineup and is positioned as the omni-modal option rather than a text-only dense model. Smaller dense Qwen3 models are cheaper to host but carry less total capacity, while larger Qwen3 checkpoints activate more parameters per token. Since we do not have verified accuracy benchmarks for this model, run your own evaluation before committing.
Is it a reasoning model?
We do not have confirmed data indicating a dedicated extended-reasoning mode for this entry. It is catalogued as a chat model, and its low active-parameter count per token suggests dedicated reasoning models may hold an advantage on math and multi-step logic tasks.