Qwen3 30B A3B Instruct 2507
Qwen · released Jul 28, 2025 · Qwen/Qwen3-30B-A3B-Instruct-2507
- Type
- Open weightsApache License 2.0
- Params
- 30.5B
- Context
- 262K
3B active per word · about 197K words of context
Our take
Written Sep 1, 2026Qwen3 is a text-only downloadable model with 30.5 billion total parameters but only 3 billion active per token, released under an Apache licence. Its coding score sits well above its overall chat quality, and twelve hosted offers give unusually wide provider choice.
Pick this for budget-conscious text inference where mixture-of-experts efficiency matters, or for high-throughput serving where one host measures 91 tokens per second. Use it for Apache-licensed local deployment that fits active parameters into small-GPU memory. Skip it if you need creative writing quality, consistent speed guarantees, or measured performance beyond 262,144 tokens.
The case for it
- Only 3 billion active parameters out of 30.5 billion total — roughly a 10:1 sparsity ratio.
- Twelve hosted offers with the cheapest costing less than half the most expensive at both input and output tiers.
- Coding Elo 1439.5, which is 56.2 points above its own overall text score.
- Apache 2.0 licence allows commercial use, fine-tuning and redistribution.
The case against it
- Overall chat quality trails its coding peak by the same 56.2 points.
- Creative writing is the weakest measured skill at 1319.6 Elo — 119.9 points below coding.
- Throughput ranges from 5 to 91 tokens per second and is unverified on several hosts.
How good is it?
EverydayGeneral questions and everyday reasoning
Arena Text (overall)109th of 168 · 1383
CodingWriting and fixing code on its own
Arena Coding101st of 168 · 1438
Arena Coding is the only board that has scored it for this.
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent.
WritingDrafting and rewriting prose
Arena Creative Writing119th of 168 · 1319
Arena Creative Writing is the only board that has scored it for this.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done.
Every published score for this model6 scoresEvery figure we hold, from 6 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
GeForce RTX 4090 · 24 GB
Borderline fit on an estimated size. It leaves 1.6 GB spare on a size we calculated rather than measured, and a 10% error either way would change the answer.
Comfortable fit
GeForce RTX 5090 · 32 GB
Room to spare. 9.6 GB spare means a 10% error in the size would not change the answer.
Apple M1 Pro (16-core GPU) · 32 GB
Room to spare. 2.8 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked between 2 hours and 21 days ago — each listing carries its own date.
- per 1M tokens
- $0.048 in / $0.19 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| StreamLakeThrough OpenRouter | $0.048 / $0.19checked 2 hours ago | 128K32K max reply | 68 tok/s | No | Yesunknown period | Unknown |
| OpenRouterOpenRouter's own listing | $0.048 / $0.19checked 2 hours ago | 262K | not measured | Unknown | Unknown | Unknown |
| DekaLLMThrough OpenRouter | $0.090 / $0.30checked 2 hours ago | 262K236K max reply | 55 tok/s | No | No | Confirmed |
| SiliconFlowfp8Through OpenRouter | $0.090 / $0.30checked 2 hours ago | 262K236K max reply | 20 tok/s | No | No | Confirmed |
| Nebius AI Studiofp8Through OpenRouter | $0.10 / $0.30checked 2 hours ago | 262K236K max reply | 59 tok/s | No | No | Confirmed |
| Novita AIDirect | $0.090 / $0.45checked 21 days ago | 41K | not measured | Unknown | Unknown | Unknown |
| DeepInfrafp8Direct and through OpenRouter | $0.12 / $0.50checked 2 hours ago | 41K16K max reply through OpenRouter | 74 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Alibaba CloudThrough OpenRouter | $0.13 / $0.52checked 2 hours ago | 131K33K max reply | 68 tok/s | No | Yesunknown period | Unknown |
Nebius AI Studio
Novita AI
DeepInfraAcross the 8 listings we hold: 6 say they do not train on prompts (1 of them only through OpenRouter), 0 say they do and 2 do not say. 4 appear in the zero-retention registry we check (1 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| StreamLakeThrough OpenRouter | ✓ | ✓ | ✓ |
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| DekaLLMThrough OpenRouter | ✓ | ✓ | ✓ |
| SiliconFlowfp8Through OpenRouter | ✓ | ✓ | ✓ |
| Nebius AI Studiofp8Through OpenRouter | ✓ | ✓ | ✓ |
| Novita AIDirect | |||
| DeepInfrafp8Direct and through OpenRouter | ✓ | ✓ | ✓ |
| Alibaba CloudThrough OpenRouter | ✓ | ✓ | ✗ |
Tool calling: 7 of 8 listings say yes, 1 publishes no parameter list. JSON output: 7 of 8 listings say yes, 1 publishes no parameter list. Strict schema: 6 of 8 listings say yes, 1 says no, 1 publishes no parameter list.
Models people weigh against Qwen3 30B A3B Instruct 2507
When we formed this view
Recent changes
What moved
input −14% ($0.14 → $0.12 per 1M tokens), output −5% ($0.55 → $0.52 per 1M tokens)What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- 1 of 8 listings publishes no parameter list, so what its API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 2 of 8 listings do not say whether they train on prompts, and 1 answers only through OpenRouter, not for its own listing.
- We hold no cached-input rate for any of its listings.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- Qwen/Qwen3-30B-A3B-Instruct-2507
- Architecture
- Mixture of experts
- Takes in, gives back
- Text in, text out
- Catalogue slug
- qwen-qwen3-30b-a3b-instruct-2507