Sonnet 5.5 makes everyday work faster
Anthropic emphasizes lower task costs, clearer writing, and practical coding.
45 stories
Models
New generations, open weights, and real limitations.
Opus 5.5 combines stronger work with lower pricing
Anthropic reports progress in complex tasks and behavioral evaluations.
GPT-6 Sol and Luna bring more choices to the new family
OpenAI positions two models around different balances of capability and cost.
Local AI: where to start
Choose the task, the hardware, and the operating boundaries before choosing the model.
North Small Translate focuses on multilingual work
Cohere releases open weights with a non-commercial license.
GPT-6 Astra opens a new OpenAI generation
The announcement spans coding, computer use, cybersecurity, and science.
Gemini 3.8 Flash extends Google’s agent-focused series
The release pairs a general model with a controlled-access cybersecurity variant.
Granite 4.2 adds reasoning at several sizes
IBM describes a family with adjustable thinking and native tool calling.
Gemini 3.7 Flash focuses on coding and agents
Google follows its previous Flash release with another workflow-oriented update.
Grok 4.6 focuses on staying with a complex task
The release emphasizes coding, research, and interactive projects.
Muse Glimmer puts Meta’s model closer to the device
The multimodal release targets local agentic applications.
Opus 5 targets stronger everyday professional work
Anthropic updates its flagship with an emphasis on efficiency.
Google splits Flash into more specialized roles
Gemini 3.6 Flash, Flash-Lite, and Flash Cyber address different agent workloads.
Inkling brings Thinking Machines into open multimodal models
Hugging Face details a release spanning text, images, and audio.
GPT-5.6 offers several balances of capability and cost
OpenAI launches Sol, Terra, and Luna for different workloads.
Sonnet 5 narrows the gap between a daily assistant and an agent
Anthropic updates planning and tool use for broadly accessible workflows.
GLM-5.2 builds for longer engineering tasks
Z.AI combines a million-token context with adjustable reasoning effort.
North Mini Code gives Cohere a coding-focused model
The open release targets agents and user-controlled deployment.
Mellum2 gives workplace tools a compact model component
JetBrains targets code and supporting operations with an open-weight release.
MiniMax M3 brings coding and multimodality together
The open-weight release targets agents with long working context.
Opus 4.8 adds more control over reasoning effort
Anthropic updates its flagship and the surrounding long-task workflow.
Command A+ gives enterprise agents an open-weight option
Cohere pairs an agent-focused model with Apache 2.0 licensing.
Gemini 3.5 puts action at the center
Google starts the family with Flash and emphasizes extended agentic work.
GPT-5.5 Instant updates everyday ChatGPT
OpenAI emphasizes clearer answers and personalization controls.
Nemotron Nano Omni brings several inputs into one model
NVIDIA combines document, image, audio, and video understanding.
DeepSeek V4 asks what long context is actually for
Hugging Face examines Pro, Flash, and the needs of long-running agents.
GPT-5.5 brings more working skills into one model
OpenAI focuses on coding, research, and analysis with tools.
Kimi K2.6 tackles longer coding assignments
Moonshot AI emphasizes sustained execution and agent swarms.
Opus 4.7 puts more emphasis on checking its work
Anthropic updates coding, vision, and multi-step execution.
Gemma 4 expands Google’s open-model family
Different sizes address different devices, budgets, and tasks.
Granite Vision uses a small model to read complex documents
IBM emphasizes tables, charts, and linked fields.
Cohere Transcribe turns speech into a separate building block
The company expands into downloadable speech recognition.
GPT-5.4 mini and nano find roles inside larger workflows
OpenAI expands its smaller-model options for fast and repeated tasks.
Mistral Small 4 combines reasoning, vision, and code
The release brings several strands of Mistral’s work into one model.
GPT-5.4 combines code, documents, and computer use
OpenAI pushes its general model toward completed professional tasks.
Gemini 3.1 Flash-Lite targets high-volume work
Google introduces a smaller option for frequently repeated tasks.
Gemini 3.1 Pro tackles questions that need more work
Google expands reasoning capabilities across consumer and developer products.
Sonnet 4.6 updates Claude’s everyday model
Anthropic emphasizes coding, computer use, and longer context.
Qwen3.5 makes multimodality part of the foundation
Alibaba’s Qwen team introduces a model aimed at visual and agentic work.
Codex-Spark bets on faster iteration
A research preview targets interactive coding with less waiting.
Opus 4.6 brings more context to extended work
Anthropic updates its flagship and introduces a million-token context beta.
GPT-5.3-Codex looks beyond the code snippet
OpenAI emphasizes sustained technical work with tools.
Kimi K2.5 connects visual understanding with agents
Moonshot AI introduces native multimodality and coordinated parallel work.
A year after DeepSeek, the open ecosystem is wider
Hugging Face looks beyond one breakout model to a changing community.
Falcon-H1-Arabic puts language first
TII’s model family focuses on Arabic rather than treating it as another checkbox.