Qwen3.8 2.4T-A95B

Language Alibaba Qwen Partly verified Released Aug 12, 2026

Qwen3.8-2.4T-A95B is an open-weight, post-trained language model in the Qwen3.8 series for coding, professional work, research, and multi-step agent tasks. It requires thinking mode for every interaction, with adjustable reasoning depth. Summary written by AI Index from Alibaba Qwen’s own pages.

  • In Input modalities: Text Out Output modalities: Text
  • Reasoning
  • Tool use
  • Structured output
  • Open weights

Prices

2 offerings · prices as listed
Alibaba Cloud Model Studio (opens in a new tab) Official International · USD qwen3.8-2.4t-a95b
Conditions/ 1M tokensInputCache writeOutput
Standard $2.00—$6.00
Explicit cache —$2.50—

Prices from Alibaba Cloud Model Studio pricing (opens in a new tab) (checked Nov 9, 2026) · qwen3.8-2.4t-a95b model (opens in a new tab) (checked Nov 8, 2026)

Alibaba Cloud Model Studio (opens in a new tab) Official China · CNY qwen3.8-2.4t-a95b
Conditions/ 1M tokensInputOutput
Standard ¥12.00¥36.00

Prices from 阿里云百炼:模型价格 (opens in a new tab) (checked Nov 9, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Max thinking budget
131.07K
Built-in tools
Web search
Licence
qwen3.8-max
Family
qwen
Model ID

Not stated on Alibaba Qwen’s pages: reasoning efforts, reasoning can be off, knowledge cutoff, aliases.

Run locally

Weights: Qwen/Qwen3.8-2.4T-A95B
  • Parameters 2.45T
  • Active 95B · 512 experts, 10 per token
  • Native precision BF16
  • Layers 92
  • KV cache per token 92 KB
  • Native context 256K (model card)
  • Config positions 256K (config.json max_position_embeddings)
  • API context 977K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Doesn’t fit IQ4_XS (automatic) 8K context
WeightsKV cacheOverheadUsable
Needs
≈ 1271.5 GB (weights 1210.3 + KV cache 0.7 + overhead 60.5)
Usable
21.6 GB (90% of its memory)
Try
it’s too large for this hardware; see the smallest that runs it below
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
BF16 (native)4556.4 GB≈ 4784.9Doesn’t fit—
Q8_02420.6 GB≈ 2542.3Doesn’t fit—
FP82278.2 GB≈ 2392.8Doesn’t fit—
Q6_K1871.0 GB≈ 1965.2Doesn’t fit—
Q5_K_M1620.4 GB≈ 1702.1Doesn’t fit—
Q5_01566.3 GB≈ 1645.3Doesn’t fit—
Q4_K_M1384.0 GB≈ 1453.9Doesn’t fit—
Q4_01281.5 GB≈ 1346.3Doesn’t fit—
IQ4_XS1210.3 GB≈ 1271.5Doesn’t fit—
Q3_K_Mquality drops1113.5 GB≈ 1169.9Doesn’t fit—
IQ3_XXSquality drops871.4 GB≈ 915.7Doesn’t fit—
Q2_Kquality drops854.3 GB≈ 897.8Doesn’t fit—
IQ2_XXSquality drops586.6 GB≈ 616.7Doesn’t fit—

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (Q4_K_M, 8K): no single listed machine; about 21 × 80 GB GPUs.

Run commands

The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.

Memory to fine-tune
Full≈ 36456.8 GBDoesn’t fit your hardware
LoRA≈ 4563.7 GBDoesn’t fit your hardware
QLoRA≈ 1288.8 GBDoesn’t fit your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-21. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us
  • EQ-Bench Creative Writing v3 1,844

    #18 of 95 listed models

  • EQ-Bench by Samuel J. Paech (eqbench.com; EQ-bench/EQ-bench-site on GitHub), MIT License. · License · as of Oct 28, 2026
  • changes we made

No changes recorded yet.

Sources

Alibaba Qwen’s official policies: data use, privacy, terms →

Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data

Alibaba Qwen · China · qwen.ai

Snapshots, the same family and other models from this provider