AIWHYS

Technology

multimodal model

2 synchronized stories carry this tag.

Frontier modelsPreview

Qwen4-Max Preview opens a two-week Token Plan evaluation window

Alibaba Cloud exposed Qwen4-Max-Preview for multimodal reasoning, coding and tool use through a fixed Token Plan window from 11 to 25 November. It supports OpenAI and Anthropic-compatible APIs, but pay-as-you-go access is disabled and the preview will be retired after the window.

Why it matters: This is a deliberately time-bounded evaluation rather than a durable production endpoint. Compatibility can simplify testing, but retirement and plan constraints make it unsuitable for a new production dependency.

Read storyOriginal source · docs.qwencloud.com
Frontier modelsPreview

Baidu serves ERNIE 5.5 Flash Preview while weights remain promised

Baidu released ERNIE 5.5 Flash Preview, a 280-billion-parameter mixture model with 14 billion active parameters and a one-million-token context. The API is live on Qianfan; Baidu says Apache-2.0 weights will follow but has not published them yet.

Why it matters: The release adds a long-context Chinese model option for coding, 3D and knowledge work, but separates usable hosted access from a future open-weight promise. Teams should evaluate the API they can call today rather than plan around unreleased files.

Read storyOriginal source · ernie.baidu.com