Qwen3.8-Max

Overview

Alibaba moved Qwen3.8-Max out of preview and into general availability on August 3, 2026, two weeks after showing up at WAIC Shanghai with nothing but a slide and a claim that it trailed only Claude Fable 5. This time the model shipped with what the preview didn't: a published benchmark table, real per-token pricing, and a production API. It's a 2.4 trillion-parameter Mixture-of-Experts model with roughly 95 billion active parameters per token, a 1-million-token context window, and native text, image, and video input.

The gap between preview and GA is the real story here. Our review of the preview build flagged three specific holes: no benchmark table, no active-parameter count, no confirmed price. Alibaba closed all three on August 3. What it hasn't closed is the open-weight promise - and that promise is now late by Alibaba's own stated timeline, which matters more for a company staking its pitch on being the first to open-source a Max-class flagship. This page covers the GA release; see Qwen3.8-Max-Preview for how the unverified version looked three weeks earlier.

Alibaba is positioning Qwen3.8-Max less as a chatbot upgrade and more as an autonomous-agent flagship, built on the architecture introduced with Qwen3.7-Max but scaled up and extended to multimodal input for the first time above the trillion-parameter mark.

Alibaba published real numbers this time - the question is whether they hold up next to the promises still pending. Source: unsplash.com

Key Specifications

Benchmark Performance

Alibaba's published table gives Qwen3.8-Max a mixed but truly competitive record against the current frontier, a sharp change from the preview's zero-evidence launch.

The pattern is consistent enough to trust directionally: Qwen3.8-Max wins on agentic computer-use and long-document tasks (OSWorld-Verified, PaperBench, IFBench) and holds its own on general terminal agentic work, but loses clearly to Fable 5 on the harder professional coding benchmark, SWE-bench Pro, by 12 points. On broad-knowledge reasoning (HLE), it finishes last of the four. That's a specific, checkable claim, not a blanket "second only to Fable 5" - a real improvement in how Alibaba is presenting this launch, even with the caveat that every number in the table is Alibaba's own vendor-run evaluation, not an independent one. For the current state of independently tracked numbers across the field, see our coding benchmarks leaderboard and computer-use leaderboard.

Alibaba's own Arena.ai citation puts Qwen3.8-Max at rank 5 on Text Arena, sandwiched between four different Claude configurations. Source: alibabacloud.com

Key Capabilities

Long-horizon autonomous coding. Alibaba's headline demo is oh-my-cli, a command-line agent framework the model built completely on its own: turning incoming requests into GitHub issues, claiming them through a state machine, writing code, running end-to-end tests, and merging its own pull requests. As of July 30 the run had produced 265 commits, 127 pull requests, and 151 issues over 16 days without human intervention. Checking the repository today, it's still running: 797 commits, 61 open issues, an Apache-2.0 license, and a commit merged 33 minutes before this page was published. That's a more convincing agentic-coding demonstration than a static benchmark score, though it's still a controlled showcase project rather than an independent test.

Research and competition tasks. Alibaba also reports Qwen3.8-Max reproducing a published paper on data selection for LLM reasoning, writing roughly 7,600 lines of code and running 33 GPU training rounds over five days to land a +2.71 point improvement on AIME24 over the original paper's method. In a separate 24-hour coding competition, the model's entry reportedly beat 458 of 526 human teams, landing in the 87th percentile.

Native multimodal agents at scale. Qwen3.8-Max processes documents past 200 pages and video past 100 hours using what Alibaba calls "video memory graphs," and pairs GUI screen operation with visual feedback loops for verifying its own output, evaluated internally against Alibaba's RecreationBench. This carries forward the multimodal push started with Qwen3.6-Max-Preview, now applied to a model an order of magnitude larger.

Pricing and Availability

Qwen3.8-Max is live now on QwenCloud at $2.00 per million input tokens and $6.00 per million output tokens, with cached input as low as $0.17/M on explicit reads. That undercuts Kimi K3's $3.00/$15.00 pricing by a wide margin and roughly matches Qwen3.7-Max's prior $2.50/$7.50 rate card, despite the jump in scale.

Access runs through the standard QwenCloud API (model ID qwen3.8-max), Alibaba Cloud Model Studio's international scope, and the Vercel AI Gateway at zero markup (alibaba/qwen3.8-max). Alibaba also shipped a day-one Anthropic Messages-compatible endpoint, which means Claude Code points at Qwen3.8-Max by changing ANTHROPIC_BASE_URL and ANTHROPIC_MODEL with no other workflow changes - the cheapest way to run a side-by-side comparison against Claude models already in an existing agent harness.

The open-weight release is the unresolved piece. Alibaba said weights for Qwen3.8-Max and a smaller Qwen3.8-27B would land on Hugging Face and ModelScope "the week of August 10." Now, one day past that window's start, no repository has appeared for either model, and no license has been named - meaning it's still an open question whether a Max-class open weight would ship under the permissive Apache-2.0 license used for smaller Qwen releases like Qwen3.6-27B, or something more restrictive. Until weights land, open_source: false is the accurate label for this page regardless of what's been promised. Track the open-source LLM leaderboard for when that changes.

Strengths and Weaknesses

Strengths

• Real, checkable benchmark table replaces the preview's unverified "second only to Fable 5" marketing line
• Leads the four-way comparison on OSWorld-Verified, PaperBench, and IFBench
• $2.00/$6.00 pricing undercuts Kimi K3 by a wide margin while adding native multimodal input
• Day-one Anthropic-compatible endpoint makes it a drop-in swap for Claude Code and similar agent harnesses
• The oh-my-cli autonomous coding project is a live, ongoing, publicly auditable demonstration rather than a one-time benchmark run

Weaknesses

• Trails Claude Fable 5 by 12 points on SWE-bench Pro, the harder of the two coding benchmarks in Alibaba's own table
• Finishes last of four flagships on HLE, nearly 10 points behind Fable 5
• Every benchmark number is vendor-run by Alibaba; no independent evaluator (Artificial Analysis, LMArena third-party listing) has copied the full table yet
• Promised open weights for "week of August 10" haven't shipped as of August 11, with no license confirmed
• Active-parameter figure (95B) comes from third-party reporting, not an Alibaba-published technical report or model card

Related Coverage

• Qwen3.8-Max-Preview model profile - the unverified preview build this page supersedes
• Qwen3.8-Max-Preview Review - hands-on testing of the preview, before real benchmarks existed
• Qwen3.7-Max model profile - the direct predecessor and prior production flagship
• Kimi K3 model profile - Moonshot AI's 2.8T open-weight rival that prompted Alibaba's preview timing
• Claude Fable 5 model profile - the model Qwen3.8-Max still trails on harder coding benchmarks
• Coding benchmarks leaderboard - independent tracking as third-party numbers arrive
• Open source LLM leaderboard - where Qwen3.8-Max lands once weights actually ship

FAQ

Is Qwen3.8-Max open source?

Not yet. Alibaba promised open weights for Qwen3.8-Max and a smaller Qwen3.8-27B during "the week of August 10, 2026," but as of August 11 neither model has appeared on Hugging Face or ModelScope, and no license has been confirmed.

How much does Qwen3.8-Max cost?

$2.00 per million input tokens and $6.00 per million output tokens on QwenCloud, with cached reads as low as $0.17/M. That's roughly one-third Kimi K3's per-token cost.

Does Qwen3.8-Max actually beat GPT-5.6 and Claude Fable 5?

It depends on the task. Alibaba's own table shows it ahead on OSWorld-Verified, PaperBench, and IFBench, but behind Fable 5 by 12 points on SWE-bench Pro and behind all three rivals on HLE. It's not a clean sweep in either direction.

How many parameters does Qwen3.8-Max have active per token?

Third-party reporting puts it at roughly 95 billion active parameters out of 2.4 trillion total, though Alibaba hasn't published this figure directly in an official technical report.

Can I use Qwen3.8-Max with Claude Code?

Yes. Alibaba shipped an Anthropic Messages-compatible endpoint on day one. Pointing ANTHROPIC_BASE_URL at Alibaba's DashScope endpoint and setting ANTHROPIC_MODEL to qwen3.8-max runs existing Claude Code workflows against it with no other changes.

What is oh-my-cli?

An open-source CLI coding agent framework that Qwen3.8-Max built and continues to maintain autonomously, without human commits. As of this writing the GitHub repository shows 797 commits and ongoing activity.

Sources:

• Alibaba Unveils Qwen3.8-Max: Its Largest and Most Capable Flagship Model to Date - Alibaba Cloud Press Room
• Qwen3.8-Max: A New Bar for Coding and Cowork - Alibaba Cloud Community
• qwen-code-dev-bot/oh-my-cli - GitHub
• Alibaba Qwen Releases Qwen3.8-Max: A 2.4 Trillion Parameter MoE Model and the Most Capable One in the Qwen Family to Date - MarkTechPost
• Qwen 3.8 Benchmarks: What Alibaba's Table Shows, and What It Doesn't - Apidog
• Qwen 3.8 Max Ships: 2.4T MoE, 1M Context, $2/$6 per MTok, Open Weights Next Week - Developers Digest
• Qwen3.8 Open Weights: Check This Before Downloading - Digital Applied
• Alibaba's open-weight Qwen3.8-Max takes on long-horizon AI tasks with 2.4 trillion parameters - The Decoder
• Alibaba Qwen3.8-Max claims 16-day autonomous coding run - Developer Tech