NEWS & INSIGHTS

Updates & model insights

Follow platform announcements, model launches, and capability updates from official sources.

MiniMax Releases M3.1 Flash: 3x Performance Boost, 70% Cost Reduction

On the evening of October 19, MiniMax officially released its next-generation Flash model, M3.1 Flash. This model achieves significant improvements in complex Agent tasks, computer operations, software development, and cybersecurity, with online inference speed increasing by nearly 3 times and usage costs reduced by 70%. M3.1 Flash performed excellently in benchmarks such as AutomationBench-AA and DeepSWE v1.1, with API prices significantly lowered; input and output unit prices decreased by more than 50%, aiming to support complex workflows at a lower cost.

2024-10-19Read more

LATEST UPDATES

Latest updates

· 深度求索 (DeepSeek)

DeepSeek V4.1 Flash Officially Released: Native Multimodal, Outperforms V4 Pro with Lower API Prices

DeepSeek released the DeepSeek V4.1 Flash model on September 10, 2026. This model features a 552B parameter MoE architecture with a Causal-Encoder-Decoder structure, offering native multimodal visual understanding capabilities. Benchmark tests show its intelligence level surpasses V4 Pro across the board, while significantly reducing KV Cache requirements and usage costs. The official API pricing has been lowered simultaneously, and plans are in place to gradually phase out the V4 Pro model.

· 阿里研究院

Qwen3.8-Max Upgraded to Version 0902, Enhancing Programming and Office Capabilities

Alibaba Research announced the upgrade of Qwen3.8-Max to version 0902. Building on prior open-source feedback, this version underwent further post-training tailored for programming and professional office scenarios, significantly boosting overall performance and better aligning with enterprise complex tasks, scientific research, and long-cycle workflows. The new version claimed first place on the CodeArena Frontend Programming Leaderboard, setting new records in reasoning, tool calling, and end-to-end application generation capabilities, while reducing the comprehensive cost per million tokens to $5. The model is now live on the Qwen AI Platform offering API services, and has been simultaneously integrated into Qwen Office, Qoder, and the Qwen APP.

· 智谱

Zhipu Open-Sources Native Multimodal Large Model GLM-5.3-Flash: 320B Parameters and Full-Stack Support from Domestic Chips

Zhipu officially releases and open-sources the world's first native multimodal large model GLM-5.3-Flash. The model features a total parameter count of 3200亿, with only 18B activated parameters, delivering comprehensive performance comparable to Claude Opus 4.8. Leveraging a hybrid sparse and linear attention architecture along with multiple memory optimization techniques, the model achieves significant breakthroughs in long-context processing and service overhead. Inference services are fully supported by domestic AI chip clusters, drastically reducing costs. It is now globally open-source and offers enterprise-level API access.

· MiniMax

Minimax H3 Max Video Model Now Live

The AGTCloud platform has officially integrated the Minimax H3 Max video generation model, supporting video content creation at speeds faster than real-time. Users can generate videos via text descriptions or by uploading source images. The system features high prompt adherence, accurately restoring subject characteristics, motion trajectories, and camera directions. It supports custom start and end frames to control scene progression, automatically outputting rigorously composed, well-lit, and cinematic-quality visuals to meet diverse visual production needs.

· Alibaba

Wan3.0 Release

According to official announcements from 阿里云, the video generation large model Wan3.0 officially launched today. The model has been comprehensively upgraded in dimensions such as generation duration, universal creation, comprehensive referencing, and real-world scene restoration. It supports generating videos up to 30 seconds long in a single run and adds support for inputting multiple document formats including doc, xls, ppt, pdf, and md, enhancing the fidelity of real-world visuals.

· DeepSeek

DeepSeek Launches V4-Flash-Vision-Exp Multimodal Model and Files API

DeepSeek officially announced that the experimental multimodal model DeepSeek-V4-Flash-Vision-Exp is now available on the API platform. The model maintains parity with V4-Flash in text understanding, agent scheduling, and world knowledge, while achieving significant breakthroughs in multimodal agent benchmark tests, with overall performance approaching Opus-4.8. This update simultaneously opens the image-text mixed input interface, clarifies image token billing standards, and launches the accompanying DeepSeek Harness 0.1.1 version. Additionally, the platform has officially launched the free Files API, supporting one-time upload and multiple reuse of images, effectively reducing request bandwidth consumption.

· Z.ai

GLM-5.3 Announces Safety Evaluation Progress, Open Weights Release Delayed

Z.ai announces progress on the safety evaluation of the new generation GLM-5.3; due to the model's improved capabilities in cybersecurity tasks, the release of open weights will be delayed to complete further testing and protection.

· ByteDance

Seedance 2.5 Launch: Longer Videos, Stronger Reference Control, and Targeted Editing

Seedance 2.5 is designed for continuous video creation, integrating long-shot generation, multi-material referencing, and localized controllable editing into a single workflow.

· Moonshot AI

Kimi K3 Officially Released: An Open Frontier Model with Native Vision and Million-Token Context

Kimi K3 is Moonshot AI's flagship model designed for long-horizon coding, knowledge work, and reasoning, featuring native visual understanding and a context window of up to 1M tokens.

· DeepSeek

DeepSeek V4 Pro Preview Release: Million-Token Context and Enhanced Agent Capabilities

DeepSeek V4 Pro is designed for complex reasoning, mathematics, coding, and agent tasks, with the official API offering a 1M context window and both thinking and non-thinking modes.

· ByteDance

Seedream 5.0 Lite: From Image Generator to Visual Creation Assistant

Seedream 5.0 Lite strengthens understanding, reasoning, generation, and editing through a unified multimodal architecture, and adds web search enhancement.

Platform Announcements · MiniMax