Qwen3.8.2 4T A95B
Qwen3.8 2.4T A95B is an open-weight sparse mixture-of-experts model from Qwen and the open-weight variant of [Qwen3.8 Max](/qwen/qwen3.8-max), with 95 billion active parameters out of 2.4 trillion total.
Qwen3.8-2.4T-A95BReleased 12 Aug 2026 (Aug 2026)Pricing
List prices at DeepInfra, in US dollars per million tokens.
| Tier | Input / 1M | Output / 1M | |
|---|---|---|---|
| Standard | $2.00 | $6.00 | |
| Cached input 90% off input | $0.20 | 90% off input |
What it costs a month
At DeepInfra list prices, 30 days.
- Chat assistant5,000 requests a day, 2K in / 0.5K out, cached$888
- RAG over documents2,000 requests a day, 8K in / 0.6K out, cached$744
- Coding agent1,000 requests a day, 40K in / 2K out, cached$1,248
- Overnight batch20,000 requests a day, 4K in / 0.4K out$6,240
Where to run it
4 hosts serve Qwen3.8.2 4T A95B. The comparison is on blended price, 3 parts input to 1 part output.
| Host | In / out per 1M | Model id | |
|---|---|---|---|
StreamLakeCloud qwen3.8-2.4t-a95b | $2.00 / $6.00 | qwen3.8-2.4t-a95b | |
DeepInfraInference Qwen3.8-2.4T-A95B | $2.00 / $6.00 | Qwen3.8-2.4T-A95B | |
Together AIInference Qwen3.8-2.4T-A95B | $2.00 / $6.00 | Qwen3.8-2.4T-A95B | |
AIHubMixGateway qwen3.8-2.4t-a95b | $2.00 / $6.00 | qwen3.8-2.4t-a95b |
Capabilities
Core
- Streaming
- Function calling
- JSON mode
- Structured outputs
Multimodal
- Vision
- Image input
- Audio
- Image generation
- Video
Advanced
- Reasoning
- Tool calling
- MCP
- Prompt caching
- Batch API
- Embeddings
- Takes in
- Text
- Produces
- Text
- Features
- Prompt caching
- Open weights
- License: Apache-2.0Runs in about 5870.8 GB of GPU memoryRuntimes: Ollama, SGLang, llama.cpp, vLLM
Cheaper alternatives
Current models that cost less and keep most of what this one can do: tools, vision, reasoning, structured output, caching and enough context.
Same model, cheaper host
No host sells it for less than DeepInfra.
Smaller Alibaba Qwen models
No cheaper Alibaba Qwen model keeps its key capabilities.
Other frontier labs
- 33% less
- 50% less
- 58% less
- 93% less
Open-weight picks
- 29% less
- 43% less
- 71% less
- 74% less
Call it
Using DeepInfra's API.
from litellm import completion
response = completion(
model="deepinfra/Qwen3.8-2.4T-A95B",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)Quick answers
How much does Qwen3.8.2 4T A95B cost?
On DeepInfra, Qwen3.8.2 4T A95B costs $2.00 per million input tokens and $6.00 per million output tokens. Cached input costs $0.20 per million tokens.
Which APIs offer Qwen3.8.2 4T A95B?
Qwen3.8.2 4T A95B runs on 4 APIs: StreamLake, DeepInfra, Together AI and AIHubMix.
What is Qwen3.8.2 4T A95B's context window?
262K tokens, with up to 262K tokens in one reply.
Is Qwen3.8.2 4T A95B being retired?
No retirement date is scheduled for it on any API we track.
What are cheaper alternatives to Qwen3.8.2 4T A95B?
Models that cost less and keep most of what it can do: GPT-6.1 Terra on OpenAI (33% less), Gemini 3.8 Flash on Google AI Studio (50% less) and Qwen4 Max on Together AI (29% less). Savings are on a mix of 3 input tokens to 1 output token.
When was Qwen3.8.2 4T A95B released?
Alibaba Qwen released it on 12 Aug 2026.
What is Qwen3.8.2 4T A95B's API model id?
Use Qwen3.8-2.4T-A95B with DeepInfra. Other hosts use their own ids: qwen3.8-2.4t-a95b on StreamLake, qwen3.8-2.4t-a95b on AIHubMix.
Sources
Raw catalog entry for Qwen3.8-2.4T-A95B on DeepInfra
{
"max_output_tokens": 262144,
"max_input_tokens": 262144,
"input_cost_per_token": 0.000002,
"output_cost_per_token": 0.000006,
"cache_read_input_token_cost": 2e-7,
"supports_prompt_caching": true,
"litellm_provider": "deepinfra",
"mode": "chat",
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": false,
"source": "https://deepinfra.com/pricing"
}