Qwen3-Coder-480B-A35B-Instruct
By Qwen. 480.2 billion parameters, a context window of 262,144 tokens and the licence apache-2.0.
Facts
- Released
- 2025-07-22 the day the repository was first published on Hugging Face Hugging Face, read
- Licence
- apache-2.0 the licence the model card declares Hugging Face, read
- Open weights
- Yes the weights are published in this Hugging Face repository Hugging Face, read
- Parameters
- 480.2 billion counted from the safetensors weight files Hugging Face, read
- Active parameters
- 35.5 billion estimate estimated from the expert counts and sizes in the configuration model configuration, read
- Knowledge cutoff
- Unknown
- Context window
- 262,144 tokens the maximum position embeddings in the model configuration model configuration, read
- Longest output
- Unknown
- Tool calling
- Yes the chat template in the tokenizer configuration accepts tool definitions Hugging Face, read
- Structured output
- Unknown
- Reasoning controls
- Unknown
- Inputs and outputs
- text in, text out Hugging Face, read
- Good for
- Coding: the maker's name for the model marks it for code Hugging Face, read
Prices
| Provider | Input | Output | Context | As of | Source |
|---|---|---|---|---|---|
| Nvidia | 0 USD | 0 USD | 262,144 | older than 30 days; check the provider | models.dev, read |
| Merge Gateway | 0.220 USD | 1.80 USD | 131,072 | older than 30 days; check the provider | models.dev, read |
| SiliconFlow | 0.250 USD | 1.00 USD | 262,000 | older than 30 days; check the provider | models.dev, read |
| SiliconFlow (China) | 0.250 USD | 1.00 USD | 262,000 | older than 30 days; check the provider | models.dev, read |
| Abacus | 0.290 USD | 1.20 USD | 262,144 | older than 30 days; check the provider | models.dev, read |
| Jiekou.AI | 0.290 USD | 1.20 USD | 262,144 | models.dev, read | |
| Novita AI | 0.380 USD | 1.55 USD | 262,144 | LiteLLM, read | |
| Deep Infra | 0.400 USD | 1.60 USD | 262,144 | LiteLLM, read | |
| Hugging Face | 2.00 USD | 2.00 USD | 262,144 | older than 30 days; check the provider | models.dev, read |
Run it on your own hardware
Weights are the sizes of the files a source lists, or an estimate from the parameter count where none does. Memory is an estimate: weights plus KV cache plus 512 MiB and 5 percent of the weights for runtime buffers. Check it against your hardware.
| Quantization | Weights | Memory at 8,192 tokens | Memory at 32,768 tokens |
|---|---|---|---|
| UD-IQ1_M | 139 GiB | 149 GiB | 155 GiB |
| Q2_K | 163 GiB | 173 GiB | 179 GiB |
| Q2_K_L | 163 GiB | 173 GiB | 179 GiB |
| UD-Q2_K_XL | 168 GiB | 179 GiB | 185 GiB |
| UD-IQ3_XXS | 188 GiB | 200 GiB | 205 GiB |
| Q3_K_S | 193 GiB | 205 GiB | 211 GiB |
| UD-Q3_K_XL | 198 GiB | 210 GiB | 216 GiB |
| Q3_K_M | 214 GiB | 227 GiB | 232 GiB |
| IQ4_XS | 243 GiB | 257 GiB | 263 GiB |
| IQ4_NL | 252 GiB | 267 GiB | 273 GiB |
| Q4_0 | 253 GiB | 268 GiB | 274 GiB |
| Q4_K_S | 254 GiB | 269 GiB | 275 GiB |
| UD-Q4_K_XL | 257 GiB | 272 GiB | 278 GiB |
| Q4_K_M | 270 GiB | 286 GiB | 292 GiB |
| Q4_1 | 280 GiB | 296 GiB | 302 GiB |
| Q5_K_S | 308 GiB | 326 GiB | 331 GiB |
| UD-Q5_K_XL | 317 GiB | 335 GiB | 341 GiB |
| Q5_K_M | 317 GiB | 335 GiB | 341 GiB |
| Q6_K | 367 GiB | 388 GiB | 394 GiB |
| UD-Q6_K_XL | 380 GiB | 401 GiB | 407 GiB |
| Q8_0 | 475 GiB | 502 GiB | 507 GiB |
| UD-Q8_K_XL | 511 GiB | 539 GiB | 544 GiB |
| BF16 | 894 GiB | 942 GiB | 947 GiB |
Start it with a local runtime
- Ollama
ollama run hf.co/unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q4_0 - LM Studio
lms get unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF lms server start - llama.cpp server
llama-server -hf unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q4_0 --jinja - vLLM
vllm serve Qwen/Qwen3-Coder-480B-A35B-Instruct - SGLang
sglang serve --model-path Qwen/Qwen3-Coder-480B-A35B-Instruct --port 30000
Use it from your harness
Set up for Abacus with the model Qwen/Qwen3-Coder-480B-A35B-Instruct. Each endpoint page has the same setup for its own address.
OpenCode
Put this in opencode.json in your project folder:
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"abacus": {
"npm": "@ai-sdk/openai-compatible",
"name": "Abacus",
"options": {
"baseURL": "https://routellm.abacus.ai/v1",
"apiKey": "{env:ABACUS_API_KEY}"
},
"models": {
"Qwen/Qwen3-Coder-480B-A35B-Instruct": {
"name": "Qwen3-Coder-480B-A35B-Instruct"
}
}
}
}
}- OpenCode reads any OpenAI-compatible address through the @ai-sdk/openai-compatible package, and an address that speaks the Responses API through @ai-sdk/openai.
From OpenCode documentation, read .
Pi
Put this in ~/.pi/agent/models.json:
{
"providers": {
"abacus": {
"baseUrl": "https://routellm.abacus.ai/v1",
"api": "openai-completions",
"apiKey": "$ABACUS_API_KEY",
"models": [
{
"id": "Qwen/Qwen3-Coder-480B-A35B-Instruct"
}
]
}
}
}- The apiKey field can name an environment variable as $NAME.
From Pi documentation, read .
Codex
Codex speaks only the Responses API, and Abacus documents no Responses address. A gateway that offers one can sit in between.
- Codex speaks the Responses API only: responses is the one supported wire API of a custom provider. Ollama and LM Studio are built in and start with --oss.
From Codex documentation, read .
Claude Code
Claude Code sends Anthropic Messages requests, and Abacus documents no such address. A gateway that translates to that API can sit in between.
- Claude Code sends Anthropic Messages requests to ANTHROPIC_BASE_URL. Anthropic says it does not support routing Claude Code to models other than Claude through any gateway, so some features may not work with another model.
From Claude Code documentation, read .
Other ways to reach it
- DeepInfra, model
Qwen/Qwen3-Coder-480B-A35B-Instruct - Hugging Face, model
Qwen/Qwen3-Coder-480B-A35B-Instruct - Jiekou.AI, model
qwen/qwen3-coder-480b-a35b-instruct - Novita AI, model
qwen/qwen3-coder-480b-a35b-instruct - Nvidia, model
qwen/qwen3-coder-480b-a35b-instruct - SiliconFlow, model
Qwen/Qwen3-Coder-480B-A35B-Instruct - SiliconFlow (China), model
Qwen/Qwen3-Coder-480B-A35B-Instruct - Ollama, model
hf.co/unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q4_0
Published results
These are results other people published. Baltor did not run them and does not rank models by them.
- Results the maker published on the model card published by Qwen Hugging Face, read
Two more publishers are linked, never copied, because their terms do not allow it:
- Artificial Analysis publishes its own intelligence, speed and price measurements.
- OpenRouter lists current prices and uptime for each provider it routes to.
Sources of this page
- huggingface.co/api/models/Qwen/Qwen3-Coder-480B-A35B-Instruct, read
- huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/raw/main/config.json, read
- huggingface.co/api/models/unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF/tree/main, read
- models.dev/api.json, read
- github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json, read
Paid links
No link in this directory is a paid link or an ad, and no listing is paid for. The order and the contents of every list come from the sources named on this page.