Flash

The news, the context, and what changes.

RSS
Flash

Meta Launches Lightweight AI Model Muse Ember

Meta has released Muse Ember, its smallest model to date, focusing on complex reasoning and agent tasks with support for one million token context and native multimodality. The cost per single task is approximately $0.14, about one-third of Claude Haiku 5.5. While it outperforms the latter in several agent benchmarks, it still lags in real-world engineering coding.

88Z AI editorial1 min read
Flash

Baidu Launches Latest Foundation Model ERNIE-5.5-Flash-Base

Baidu has released the ERNIE-5.5-Flash-Base foundation model, making its pre-training endpoint weights available. This sparse MoE model features 280B total parameters with 14B activated per token, pre-trained from scratch on 33.5T tokens. It supports a 128K training context and up to 1M for fine-tuning and inference. The weights are unprocessed by mid or post-training, making them suitable for domain adaptation, instruction tuning, and reinforcement learning.

88Z AI editorial1 min read
Flash

Meituan LongCat Team Open Sources LongCat-2.5

Meituan's LongCat team has open-sourced the LongCat-2.5 model under the MIT license, allowing commercial use. With 1.6 trillion total parameters and approximately 48 billion activated, it supports million-token contexts and long-horizon agent tasks. Optimized for domestic computing power, it significantly improves single-card training throughput.

88Z AI editorial1 min read
Flash

Anthropic Launches Flagship Models Claude Fable 5.2 and Claude Mythos 5.2

Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, demonstrating strong autonomous collaboration and specialized problem-solving capabilities. A thousand agents built a bootable operating system in eight days, completed the formal proof of Fermat's Last Theorem in thirty hours, designed a four-layer PCB and manufactured three real units in one day, and cracked viral capsid folding in seven minutes on a single card. Benchmark performance is excellent, with Fable 5.2 offering equivalen

88Z AI editorial1 min read
Flash

Alibaba Launches Anonymous Coding Model Qoduck, Available for Free on Qoder

Alibaba's Qoder has launched the anonymous model Qoduck, represented by a black duck icon without disclosed parameters. Praised for its speed and high success rate in long-horizon tasks and file editing, it supports complete deliverables like videos. It is currently free for individual users for a limited time.

88Z AI editorial1 min read
Flash

ByteDance Seed Team Launches Image Creation Model Seedream 5.5

ByteDance's Seed team has released Seedream 5.5, a new image creation model now available on Jimeng and Doubao. It supports up to 100 rounds of continuous editing with pixel-level preservation of unchanged areas, generates native RGBA transparent assets, and reduces text distortion to 0.4%, enhancing realism and design workflow efficiency.

88Z AI editorial1 min read
Flash

Google Launches Latest Video Generation Model Gemini Omni 1.2 Flash

Google has released the Gemini Omni 1.2 Flash video generation model with full API access. It supports native 15-second video generation, with prices reduced by 30% compared to version 1.1, costing only $0.07 per second for 720p resolution. Upgrades include multi-turn conversational editing, strong first-and-last frame consistency, reference image and audio guidance, more natural lip-syncing, and cinematic camera control.

88Z AI editorial1 min read
Flash

OpenAI Launches Cybersecurity-Specific Model GPT-6 Cyber

OpenAI has officially launched GPT-6 Cyber, a cybersecurity-specific model capable of autonomously discovering and exploiting vulnerabilities. The model achieved a 48.5% success rate in ExploitGym tests, an approximately 51% improvement over the previous generation. Furthermore, it reproduced 1,724 vulnerabilities in the Linux kernel, generating 948 usable exploits, and identified 81 unknown high-risk vulnerabilities in Firefox, some of which have been patched by Mozilla.

88Z AI editorial1 min read
Flash

Baidu Open-Sources Latest Base Model ERNIE-5.5-Flash-Base

Baidu has officially open-sourced the ERNIE-5.5-Flash-Base pre-trained base model, with weights now available on Hugging Face. The model offers FP8 and BF16 precision versions, is compatible with SGLang and vLLM deployment frameworks, and supports fine-tuning via Megatron-Bridge. As a sparse MoE architecture model, it features 280 billion total parameters with 14 billion activated per token, pre-trained from scratch on 33.5 trillion tokens, and supports a context window of up to 1 million tokens

88Z AI editorial1 min read
Flash

Ant Bailin Launches Native Multimodal Model Ling-3.2-flash

Ant Bailin has released Ling-3.2-flash, a native multimodal model with 560B total parameters and 25B activated per step, supporting a 1M context window. It features visual feedback self-check and adaptive anomaly recovery, allowing it to correct biases using rendering or interface results and retry after errors or tool failures. Using a hybrid attention mechanism of 7-layer KDA and 1-layer Gated MLA, it achieves an average decoding speed of 280 tokens/s for single users with long contexts under

88Z AI editorial1 min read
Flash

Kling AI 4.0 Flash Now Fully Available

Kling 4.0 Flash is now fully available, supporting access via the official website, API, and various development tools, while also compatible with Adobe software plugins to seamlessly integrate video generation into post-production workflows.

88Z AI editorial1 min read