whatmodel.sh / Qwen3-Coder 480B-A35B Instruct
Qwen3-Coder 480B-A35B Instruct
↗ share
+ contribute
Qwen3-Coder 480B-A35B Instruct is an open-weights language model from Alibaba, released July 2025. Its published size is 480B parameters (35B active per token) and a 262k-token context window. Licence: apache-2.0.
$ how it compares
The LLMs ranked either side of Qwen3-Coder 480B-A35B Instruct, out of 828 with a score, then more from Alibaba.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 177 |  Nemotron-Labs-3 Puzzle 75B-A9B | 64±1 | local | ||
| 178 | Qwen3-VL 235B-A22B Thinking | 64±1 | local | ||
| 179 | Qwen3-Coder 480B-A35B Instruct | 64±1 | api $0.67/M +local | ||
| 180 |  Mistral Medium 3.1 (2508) | 64±2 | comparison only | ||
| 181 |  Gemma 4 26B-A4B | 64±1 | api $0.14/M +local | ||
| more from Alibaba | |||||
| 18 | Qwen3.8 Max (2.4T-A95B) | 78±1 | api $3.00/M +local | ||
| 34 | Qwen3.8 Max (0902) | 76±1 | api $3.00/M | ||
| 37 | Qwen3.8-Flash-Next | 75±1 | local | ||
$ what it's good at
agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #132 of 300 · 36% of its score · 7 tests
Beats o3, Gemini 2.5 Pro, DeepSeek V3.2 Exp here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Large, GPT-4o (2024-05-13) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Terminal-Bench 2.0 with Terminus 2 agent (benchmark authors' runs) | 23.9% ✎ | ▃··▃·▃▃██▅▃▅ beats 4/14 | 11% |
| SWE-bench Verified lab +1 | 67.0% ✎ | ▁▁▁▁▁▁▂▄█▆▁▁ beats 74/131 | 8.0% |
| τ-bench Retail lab +1 | 77.5% ✎ | ▁····▃·▂▇█▂▆ beats 23/28 | 5.9% |
| SWE-bench Multilingual lab +1 | 54.7% ✎ | ▁··▁▃▂▅▇█▃▃▁ beats 15/60 | 5.2% |
| τ-bench Airline lab +1 | 60.0% ✎ | ▁▁▁·▅▄▁█▇▁▁▇ beats 26/28 | 4.6% |
| BFCL v3 lab +1 | 68.7% ✎ | ▁▁▁▁▁▃▅█▆▅▄▁ beats 143/170 | 1.6% |
| SWE-bench Live lab +1 | 26.3% ✎ | █··········█ beats 1/1 | — |
coding▁▁▁▃▆██▇▆▄▃▁ #141 of 458 · 63% of its score · 6 tests
Beats Gemini 2.5 Pro, Gemini 3.1 Flash-Lite, Qwen3-Coder-Next (80B-A3B) here, though they rank higher overall. Trails Devstral Small 2 24B, Llama 3.1 Nemotron Ultra 253B v1, Nemotron 3.5 Lightning 30B-A3B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena WebDev (Code Arena) | 1274.2 ✎ | ▁▂▄▃▅█▄▇▆▄▁▁ beats 18/119 | 24% |
| Aider polyglot (225 exercises, two attempts) lab +1 | 61.8% ✎ | ▂▁▁▃▆▃▄▆█▃▄▂ beats 55/71 | 20% |
| SWE-bench Verified lab +1 | 67.0% ✎ | ▁▁▁▁▁▁▂▄█▆▁▁ beats 74/131 | 8.0% |
| WebApp1K-Duo-React pass@1 (onekq-ai) | 61.2% ✎ | ▁··▁·▁▁▁▆█▇█ beats 15/45 | 6.5% |
| SWE-bench Multilingual lab +1 | 54.7% ✎ | ▁··▁▃▂▅▇█▃▃▁ beats 15/60 | 5.2% |
| SWE-bench Live lab +1 | 26.3% ✎ | █··········█ beats 1/1 | — |
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #122 of 393 * · 13% of its score · 1 test
Beats LongCat 2.0 (1.6T-A48B), MiMo-V2-Flash (309B-A15B), MiniMax-M2.1 here, though they rank higher overall. Trails LFM2.5 2.6B, Nemotron-Cascade-8B, DeepSeek V3 0324 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena (text) | 1387.3 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 228/379 | 13% |
$ where to run it
Through an API
api · Novita AI · $0.67/M · $0.38 in / $1.55 out per 1M tokens · 262k context · novita.ai ↗ · price sourceUSD per 1M text tokens, standard uncached input and output. Model endpoint qwen/qwen3-coder-480b-a35b-instruct. Product: Serverless chat completions. API model: qwen/qwen3-coder-480b-a35b-instruct. Settings: service tier=standard, batch=no, cache=uncached input. Additional/excluded charges: Tools, cache storage, priority service and other modalities excluded.. Currency: USD.
llama.cpp · recommended
version
- Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.macOS / Linux
curl -LsSf https://llama.app/install.sh | shWindowsirm https://llama.app/install.ps1 | iex - Run Qwen3-Coder 480B-A35B Instruct · needs ~144.5 GB · from huggingface.co/unslothchat in the terminal
llama cli -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:UD-Q2_K_XL - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:UD-Q2_K_XL
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run Qwen3-Coder 480B-A35B Instruct · needs ~270.1 GB, Mac only · from huggingface.co/lmstudio-community
- Or download it in the terminal
lms get https://huggingface.co/lmstudio-community/Qwen3-Coder-480B-A35B-Instruct-MLX-4bit -y - Then run lms ls to see its name, and lms chat followed by that name
Ollama
- Install Ollama
download it (macOS / Windows) ↗
Linuxcurl -fsSL https://ollama.com/install.sh | sh - Run Qwen3-Coder 480B-A35B Instruct · needs ~288.5 GB · from ollama.comchat in the terminal
ollama run qwen3-coder:480b
MLX (Mac)
version
- Install MLX (Mac) · Apple silicon Macs only, text only.with uv
uv tool install mlx-lmorpip install mlx-lm - Run Qwen3-Coder 480B-A35B Instruct · needs ~270.1 GB, Mac only · from huggingface.co/lmstudio-communitychat in the terminal
mlx_lm.chat --model lmstudio-community/Qwen3-Coder-480B-A35B-Instruct-MLX-4bit - Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model lmstudio-community/Qwen3-Coder-480B-A35B-Instruct-MLX-4bit
Official weights and documentation: huggingface.co/Qwen ↗
$ how the score was worked out · Qwen/Qwen3-Coder-30B-A3B-Inst… 45%, LMArena leaderboard dataset 37%, Terminal-Bench: Benchmarking… 11% and 1 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Qwen/Qwen3-Coder-30B-A3B-Instruct model card · lab-reported | 6 | 45% |
 LMArena leaderboard dataset | 2 | 37% |
 Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces — benchmark authors' own agent-model trials | 1 | 11% |
| WebApp1K models leaderboard (onekq-ai) | 1 | 7% |
| prior · pull toward the middle | 1% | |
| weighted average | = 64.0 |
not counted (8)
| Qwen/Qwen3-Coder-480B-A35B-Instruct model card | Aider polyglot (225 exercises, two attempts) [correct] | 61.8% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| Qwen/Qwen3-Coder-480B-A35B-Instruct model card | BFCL v3 [correct] | 68.7% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| Qwen/Qwen3-Coder-480B-A35B-Instruct model card | SWE-bench Live [correct] | 26.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| Qwen/Qwen3-Coder-480B-A35B-Instruct model card | SWE-bench Multilingual [correct] | 54.7% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| Qwen/Qwen3-Coder-480B-A35B-Instruct model card | SWE-bench Verified [correct] | 69.6% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| Qwen/Qwen3-Coder-480B-A35B-Instruct model card | τ-bench Airline [correct] | 60.0% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| Qwen/Qwen3-Coder-480B-A35B-Instruct model card | τ-bench Retail [correct] | 77.5% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| Qwen/Qwen3-Coder-30B-A3B-Instruct model card | SWE-bench Live [correct] | 26.3% | fewer than three models have this test yet, or this source ran it for this model alone |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json