Qwen3.8 2.4T-A95B

Flagship

Qwen3.8 2.4T-A95B is the largest open-weight language model published to date and the downloadable counterpart to Alibaba's hosted Qwen3.8-Max. It is a fine-grained mixture of experts: 2.4 trillion total parameters with 95 billion active per token, 92 layers, and 512 experts of which 10 route plus 1 shared. The model card reports 92.6 on GPQA Diamond, 86.6 on Terminal-Bench 2.1, 67.7 on SWE-bench Pro, and 93.0 on PaperBench. It is text only, with thinking mode permanently enabled. Two details deserve attention before you commit. The license is a custom Qwen3.8-Max License, not the Apache 2.0 that covers the smaller Qwen3.8 27B, so read it before commercial deployment. And the native context window is 262,144 tokens, extensible to roughly 1M with YaRN; the larger figure quoted by hosted providers is the extended one. Self-hosting this is a multi-node proposition, roughly 2.5TB of accelerator memory even at FP8.

Input Price

$2.00

per 1M tokens

Output Price

$6.00

per 1M tokens

Context Window

1.0M

tokens

Released

2026-08

Open source

Capabilities

texttool-usecodereasoning

Key Strengths

✓Largest open-weight model published, 2.4T total with 95B active
✓92.6 GPQA Diamond and 67.7 SWE-bench Pro (vendor-reported)
✓262K native context, extensible to roughly 1M with YaRN
✓Downloadable weights for a model at genuine frontier scale
✓Fine-grained MoE keeps active compute at 4 percent of total

Best For

▸Frontier reasoning where weights must stay in-house
▸Research on very large mixture-of-experts models
▸Long-horizon agentic coding
▸Sovereign or air-gapped deployments at rack scale

Pricing Details

Input tokens

$2.00

per 1M tokens

Output tokens

$6.00

per 1M tokens

Estimated cost per 1K requests

$5.00

~1K input + ~500 output tokens avg

Prices are subject to change. Check the official documentation for current pricing. See the cost calculator for detailed estimates.

Open Source Model

Qwen3.8 2.4T-A95B is free to download and self-host under the Qwen3.8-Max License. Hosted API pricing varies by provider (e.g., Together, Fireworks, Groq). See our open source LLM guide for deployment options.