Qwen3-4B-Instruct-2507
An updated 4B-parameter non-thinking causal language model from Qwen, with significant improvements in instruction following, logical reasoning, text comprehension, mathematics, science, coding and tool usage.
text → text · made by Alibaba
Sold by 2 ways
| Seller | Lane | Rate |
|---|---|---|
| Hugging Face Inference Providers aggregator | nscale | $0.01per Mtok in$0.03per Mtok outrouter.huggingface.co · read 2026-08-25 |
| Fireworks AI aggregator | standard | $0.2per Mtok in$0.2per Mtok outraw.githubusercontent.com · read 2026-09-16 |
Measured 3 standings
| Place | Board | Metric | Score |
|---|---|---|---|
| 166thof 222 | Chess puzzles — Epoch AI | mean_score | 4 |
| 179thof 290 | OTIS Mock AIME 2024-2025 | mean_score | 52.222 |
| 252ndof 313 | GPQA Diamond | mean_score | 45.833 |
About
Qwen3-4B-Instruct-2507 — a text model from Alibaba, sold by 2 companies from $0.2 in and $0.2 out per million tokens, placed 179th of 290 on OTIS Mock AIME 2024-2025.
It takes text and returns text, with a context window of 262,144 tokens. It was published in July 2025. Its sellers say it can reason step by step and call a tool. The catalogue files it under reasoning. Two companies sell it. The cheapest is $0.2 in and $0.2 out per million tokens at Fireworks AI. Beside the standard rate there is a separately routed lane. It has been measured on 3 boards, and stands best at 179th of 290 on OTIS Mock AIME 2024-2025.
Every current figure
- Maker
- Alibaba
- Register
- model
- Takes
- text
- Returns
- text
- Context
- 262,144 tokens
- Longest answer
- 32,768 tokens
- Published
- July 2025
- Parameters
- 4 billion · read from its own name
- Licence
- not read
- Sellers
- 2
- Maker's own price
- not read
- Price
- $0.2 in and $0.2 out per million tokens — Fireworks AI
- Boards
- 3
- Best place
- 179th of 290 — OTIS Mock AIME 2024-2025
Known as 1 name
Qwen/Qwen3-4B-Instruct-2507