Chat with Qwen3.8 2.4T.

The open version of Alibaba’s biggest Qwen, 2.4 trillion parameters in all with 95B awake for each word.

Start chatting with Qwen3.8 2.4T now — free, no account

At a glance

  • Maker: Alibaba
  • Released: 2026
  • Parameters: 2.4T
  • Active per token: 95B
  • Context window: 1M
  • Size: Very large
  • Memory: Handles very long documents
  • Good at: Heavyweight reasoning
  • Supports: tools, zero data retention
  • License: Apache 2.0
  • Upstream API price: $6.00 per million tokens

What is Qwen3.8 2.4T?

Qwen3.8 2.4T is the open version of the biggest Qwen Alibaba builds — 2.4 trillion parameters, the heavyweight of one of the world's most prolific open-model families. Where most of the Qwen3.8 generation is built for volume and speed, this one is built for weight: Alibaba's own summary is heavyweight reasoning, and the model exists to compete with the largest open flagships from Moonshot, DeepSeek and Z.ai.

The architecture is mixture-of-experts at grand scale: 95 billion parameters awake per token out of 2.4 trillion total — the a95b in the slug. That is more active compute per token than many entire models on this site possess, which is what heavyweight reasoning costs. The context window is a full million tokens, roughly 750,000 words; it supports tools and zero-data-retention terms; and it is text-only, with picture input a job Alibaba leaves to the Flash tier.

At $6.00 per million tokens it is the second-priciest model on UsingOpen, behind only Kimi K3, and the positioning is deliberate: this is the Qwen you rent when the question is hard enough to justify it. For everything else, the same family sells speed at a fraction of the price.

What can you do with Qwen3.8 2.4T?

What 2.4 trillion parameters are for:

  • The hardest reasoning problems: multi-constraint planning, competition-grade mathematics, stubborn logic.
  • Deep coding work where subtle bugs and architecture questions resist smaller models.
  • Very long analyses over entire document collections — roughly 750,000 words in one window.
  • Orchestrating and reviewing work done by cheaper Qwen models in a pipeline.
  • Research and evaluation work that needs a top-tier open reference model.
  • Tasks where you would otherwise say: this needs the biggest model I can get.

How does Qwen3.8 2.4T compare?

Its only true rival at this scale is Moonshot's Kimi K3 — 2.8T at $8.50, with picture and video input versus the Qwen's text-only, and a harder coding focus. The Qwen is a little cheaper and a little smaller; both are open heavyweights aimed at heavyweight reasoning, and together they set the ceiling on what open spending buys here.

Inside its own family the 2.4T is the ceiling: Flash and the 27B do the volume work at a fraction of the price, and escalating to this model is a deliberate act — the Qwen you pick when the answer really matters.

Is Qwen3.8 2.4T right for you?

The 2.4T is the Qwen you reach for when the question has already embarrassed two smaller models: competition-grade reasoning, deep architectural analysis, the class of problem where you want the biggest open heavyweight Alibaba ships. At $6.00 per million tokens it is the second-priciest model on the site, and it behaves like it — 95 billion parameters awake per token is more active compute than many entire models here possess.

It is emphatically not a daily driver. The same family sells speed at $0.47 and care at $3.20, and using the 2.4T for ordinary drafting is paying heavyweight money for middleweight work. Its text-only input means visual material goes to the Flash tier regardless. The people the 2.4T is for know who they are: researchers who need a top open reference, teams whose escalation tier has to actually be an escalation tier.

Where it shines is the ceiling. Research that needs a reproducible frontier-class baseline, pipelines that push their hardest residue upward, evaluations measuring the distance between almost and actually — that is 2.4T territory. If you keep one model for the questions that justify 95 billion awake parameters, this is Alibaba's answer, and only Kimi K3 answers louder.

Frequently asked questions

What do the 2.4T and a95b in the model name mean?

2.4T is the total parameter count — 2.4 trillion. a95b is the active count: about 95 billion parameters are awake for any given token, a mixture-of-experts design that keeps even a model this size responsive. For comparison, the 27B sibling is dense — all 27 billion active — which is why the 2.4T can outweigh it ninety-fold in total while remaining servable.

What is the context window of Qwen3.8 2.4T?

One million tokens — roughly 750,000 words. Entire research corpora, codebases and archives fit in a single conversation. It matches Kimi K3’s window, so the heavyweight comparison between those two is purely about depth and input types rather than memory.

Does Qwen3.8 2.4T support image input?

No. It is text-only, like most of the heavyweight reasoners here. Picture and video input in the Qwen3.8 family belongs to the Flash and 27B tiers. That division of labour is deliberate across the Qwen3.8 family: the Flash and 27B tiers carry the eyes, the 2.4T carries the weight, and together they cover the range.

Why is Qwen3.8 2.4T so expensive?

Because 95 billion awake parameters per token is more active compute than many complete models on this site. At $6.00 per million tokens it costs what heavyweight reasoning costs; the same family's Flash tier exists for everyone else. For teams that only occasionally need the ceiling, the trick is to keep a cheaper Qwen as the default and treat the 2.4T as a paid escalation — the family’s shared tooling makes the hand-off easy.

Who makes Qwen3.8 2.4T?

Alibaba, through its Qwen team. This is the open release of the largest Qwen of the 3.8 generation, continuing the family's Apache-2.0 tradition. Alibaba’s open releases have made Qwen one of the most forked and fine-tuned families in existence, so community tooling for a model this size arrives almost immediately.

Is Qwen3.8 2.4T free to chat with on UsingOpen?

Yes. Pick it in the model picker and start chatting — no account, no payment, with history saved only on your device. Trying a heavyweight for free before committing budget to it is exactly what UsingOpen is for — the picker keeps it one click away.

Related models on UsingOpen

If Qwen3.8 2.4T is not quite the right shape, these neighbours on UsingOpen are worth a look:

  • Kimi K3 by Moonshot AI: the biggest model on UsingOpen — 2.8T parameters, built for hard coding.
  • Qwen3.8 Flash by Alibaba: the quick multimodal workhorse of the Qwen3.8 family, with a million-token window built for agent work.
  • GLM 5.3 by Z.ai: the full-size flagship — 753B parameters for the longest, hardest coding jobs.
  • DeepSeek V4 Pro by DeepSeek: the biggest DeepSeek — the full-strength 1.7T V4, built for hard problems.

Head-to-head comparisons

See how Qwen3.8 2.4T lines up with the models people most often weigh it against:

Model availability changes as upstream capacity comes and goes — the picker on the chat page shows what is ready right now.

Chat with Qwen3.8 2.4T free on UsingOpen — no account, no sign-up

← All open-weight models · Compare models · Free AI chat · FAQ