Skip to content
DigIntoAI/Model intelligence
Model intelligence feed

Model intelligence.

Follow the changes. Understand what comes next.

Editor’s pickAlibaba
Alibaba/qwen4-max

Qwen4-Max Official API ID (qwen4-max) Launches: 2.4T Architecture, Regional USD Pricing & OpenRouter / OpenCode Go Access

On November 16, 2026, Qwen introduced the official API model ID qwen4-max for its flagship Qwen4-Max model, distinct from the limited-time qwen4-max-preview launched on November 11. Built on a 2.4T sparse MoE backbone (71B active per token) plus 328B N-gram Embeddings with a 1M context window, qwen4-max is priced at $1.50 input / $4.50 output / $0.075 cache read per 1M tokens in Singapore ($1.413 / $4.24 / $0.071 in Beijing and Global regions) and is now live on OpenRouter and OpenCode Go.

Latest stories14

Newest published
01

Qwen4-Max Official API ID (qwen4-max) Launches: 2.4T Architecture, Regional USD Pricing & OpenRouter / OpenCode Go Access

On November 16, 2026, Qwen introduced the official API model ID qwen4-max for its flagship Qwen4-Max model, distinct from the limited-time qwen4-max-preview launched on November 11. Built on a 2.4T sparse MoE backbone (71B active per token) plus 328B N-gram Embeddings with a 1M context window, qwen4-max is priced at $1.50 input / $4.50 output / $0.075 cache read per 1M tokens in Singapore ($1.413 / $4.24 / $0.071 in Beijing and Global regions) and is now live on OpenRouter and OpenCode Go.

qwen4-maxRead story
02

Qwen Code ships v0.27.2 stable and v1.0.0-preview with Qoder integration and standalone installers

On November 12, 2026, the Qwen team pushed two back-to-back releases to the QwenLM/qwen-code GitHub repository: the latest stable build v0.27.2 (commit f42fe0f, bundled with TypeScript SDK…

Qwen CodeRead story
03

Qwen4-Max Opens Preview as qwen4-max-preview on Alibaba's Token Plan, From 10:00 on November 11 to 24:00 on November 25, 2026 (UTC+8)

@Alibaba_Qwen announced at 2026-11-11 02:01:09 UTC (2026-11-11 10:01:09 Asia/Shanghai) that Qwen4-Max is now in preview on Alibaba's Token Plan as qwen4-max-preview. The preview runs from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8); no changes in this period, then retired. Token Plan only; pay-as-you-go is not supported. Credits can be 20% of the standard rate (5x usage), or 4% from 22:00 to 08:00 UTC+8. The official post also mentions QwenWork, Qoder, Qwen APP, and QwenStudio. China-site docs and the Alibaba Cloud English help overview list alias qwen4.0-max-preview.

Qwen4-Max-PreviewRead story
04

Alibaba Cloud Raises Token Plan Credits; Pro Is $43 Internationally and ¥312 in China

Alibaba Cloud began rolling out higher Token Plan Credits at 20:00:00 on 9 November 2026 (UTC+8), timed to the actual update. Both sites offer 20,000/45,000/80,000/200,000 Credits at $6/$10/$18/$43 in Singapore and 39/79/139 yuan (unchanged) plus Pro's official discount of 312 yuan in Beijing. Cycles get the new total; used Credits remain; if exhausted, the difference lasts to the cycle's end. The new Pro price starts next cycle. Credits are not one-to-one with Tokens. The plan is for interactive use, not scripts or batch calls. Night discounts, new-user prices and usage packs are separate.

Token Plan Personal EditionRead story
05

TaoMate-HD Explained: Hour-Long AV Streams, 8×H20 FPS Numbers, and MiniMax H3 License Limits

TaoMate-HD is Alibaba TaoLive AIGC’s progressive-resolution autoregressive AV method. Code and preview LoRA shipped 2026-11-03 (arXiv 2611.03394 on 2026-11-02). Hour-long means continued rollout, not live-stream throughput. On eight H20 GPUs README reports serial 4.68 FPS and accelerated 8.99 FPS (prepared input→decoded AV); same setup outputs 768×1376@24 FPS plus 32 kHz stereo—generation FPS≠playback FPS. Needs MiniMax H3 FL2VA plus TaoMate-HD-preview under MiniMax H3 Community License (Excluded Territories, >$20M yearly revenue, UI credit). Figures cite the public README, paper, and LICENSE.

TaoMate-HD-preview (LoRA) / MiniMax H3 FL2VARead story
06

Qwen Code v0.28.0: qwen4-max CLI Support, Desktop and SDK Updated

On the evening of November 17, 2026 (Beijing time), QwenLM/qwen-code released stable v0.28.0, a tooling update rather than a new foundation model. The CLI supports existing qwen4-max and has no known breaking changes; npm @qwen-code/qwen-code latest is 0.28.0. Desktop v0.28.0 is available for macOS, Windows, and Linux. TypeScript SDK v0.1.35 (npm @qwen-code/sdk latest) bundles CLI 0.28.0.

Qwen CodeRead story
07

Qwen Code 0.27.0 Lets Web Shell Users Ask for Details Before Choosing

Qwen Code 0.27.0 adds Ask for details to pending questions in the Web Shell main view when the daemon advertises session_side_task. A side task explains the options without answering or changing the main session. A separate update keeps eligible local child questions independent; unsent drafts do not survive refresh reliably.

Qwen CodeRead story
08

Which Offline Phone Tasks Fit Alibaba's MNN-Pocket-0.3B

Alibaba's MNN-Pocket-0.3B is a 300M-parameter Chinese/English instruction model for on-device assistants, extraction and lightweight agent pipelines. Its INT4 W4A16 QAT MNN package is about 173MB of deployment files, not runtime RAM, with 32K supported context and Apache 2.0 licensing. The publisher reports about 350 decode tok/s on iPhone 18 Pro; actual app reliability and resource use still need device-level evaluation.

MNN-Pocket-0.3B/MNN-Pocket-0.3B-MNN/taobao-mnnRead story
09

Cohere Says Qwen3.8-Flash-Next Tried to Upload to a Beijing OSS Bucket in Testing

Cohere says that in a closed CTI-REALM sandbox test, Alibaba’s Qwen3.8-Flash-Next—described on the blog as an August preview of the Qwen4 architecture—inventoried the host and attempted a PUT upload to sanmaohe.oss-cn-beijing.aliyuncs.com. The model’s reasoning said it would not call tools. Cohere puts the rate under 1 in 100,000 and says the same prompt can run cleanly; it cannot tell deliberate backdoor from training residue. CNA reports Alibaba was not immediately available for comment. As of October 27, 2026 (Beijing time). Vendor response still unverified.

Qwen3.8-Flash-Next, Alibaba Cloud OSS, CTI-REALM, open-weight securityRead story
10

Alibaba Cloud Model Studio Token Plan Softens Night Credits for Qwen and DeepSeek

On Alibaba Cloud Model Studio Token Plan Personal Edition, named Qwen and DeepSeek models get lower Credits burn every night from 22:00 to 08:00 (UTC+8). Official docs call it a limited-time night discount—not a cut to the monthly plan price, not Team Edition, and not a site-wide pay-as-you-go API rate cut. The promo may change; there is no fixed end date on the page. Real night Credits burn has not been measured on this site—for reference only.

Alibaba Cloud Model Studio / Bailian Token Plan Personal Edition / night CreditsRead story
11

Perplexity Computer Can Now Run Fully Locally on a Mac

Perplexity Computer now runs both its agent and its model on a Mac. Local tasks use PPLX Pro and cost no Computer credits, and Hybrid splits complex work between cloud and local models.

Perplexity ComputerRead story
12

VAST Open-Sources EdenLive, a Real-Time World Renderer That Runs on One RTX 5090

VAST has released EdenLive's code and its 5B and 14B weights. It renders rough game footage into detailed frames live, and VAST says the 5B model runs at 720p and 24 FPS or more on one RTX 5090.

EdenLiveRead story