Skip to main content
Alibaba QwenAug 12, 2026

Qwen 3.8 2.4T

qwen-3-8-2-4t-a95bPrivateOpen weights

Qwen 3.8 2.4T is Alibaba’s open-weight 2.4-trillion-parameter MoE model (95B active), with major gains in software engineering, research, and long-horizon agentic tasks. It is text-only, requires thinking mode, and supports a 262K-token context window.

$2.50/$7.50
per 1M in / out
256K
context window
64K
max output
How it comparesamong 143 text models
GPQA Diamondhigher than 17%65.2%
Blended pricecheaper than 32%$3.75
Context windowlarger than 32%256K
Max outputlarger than 37%64K

Further right is better. Hover a dot for its model.

Time to first token— 
Output speed— 
Uptime— 

Last 24 hours of Venice traffic.How it’s measured

Capabilities

What the model takes in, returns and supports through the API.

InText
OutText
Streamingstream: true
Function callingtoolsReasoningAlways on
Web searchenable_web_search
Prompt cachingReads at $0.3125
Optimized for codeTuned for software work

Not supportedstructured outputs, image input, video input, audio input, X search and log probabilities

Benchmarks

GPQA Diamond
65.2%114th of 139 · 69 score higher for less, including Gemini 3.8 Flash, Gemini 3.7 Flash and Pareto

Specs

Model ID
qwen-3-8-2-4t-a95b
Endpoint
POST /api/v1/chat/completions
Context window
262,144 tokens
Max output
65,536 tokens
Reasoning
Always on
Precision
Not disclosed
Privacy
Private
Added
Aug 12, 2026

Compare

ModelInputOutputGPQA DiamondContextMax outputFirst tokenSpeed
Qwen 3.8 2.4T$2.50$7.5065.2%256K64K——
Grok 4.6$2.27$6.8086.4%500K200K——
GPT-6 Sol$2.00$10.0092.9%1.05M128K——
Claude Sonnet 5$2.00$10.0085.9%1M64K——
Claude Sonnet 5.5$2.00$10.0090.4%1M128K——

Pricing

Billed per token in USD. Reasoning tokens bill as output.

Input$2.50per 1M tokens
Cached input$0.312588% off input
Output$7.50per 1M tokens

Estimate a workload

$0.0088per request
$87.50for 10,000 requests

Blended price against similar models

Qwen 3.8 2.4T$3.75

Per 1M tokens at three input tokens for each output token.

Quickstart

OpenAI-compatible: point any OpenAI SDK at the Venice base URL with a Venice API key.

curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen-3-8-2-4t-a95b",
"messages": [{ "role": "user", "content": "Explain TEE attestation in two sentences." }]
}'
Base URL https://api.venice.ai/api/v1

Parameters

ParameterValuesNotes
modelqwen-3-8-2-4t-a95bRequired
messagestextContent parts the model reads
max_completion_tokensUp to 65,536Counts reasoning tokens
toolsFunction definitionsWith tool_choice
streamtrue | falseServer-sent events
venice_parameters.enable_web_searchauto | on | offWeb results and citations