AI Releases

Claude Opus 5.5 is an AI model for writing and coding from Anthropic, the company behind Claude, and the first model in its new Claude 5.5 family. Anthropic says it is especially good at long, sprawling jobs such as updating or auditing a whole codebase. Compared with Claude Opus 5, it costs less to run and writes more clearly, putting the most important information first, according to Anthropic. It is available on all of Anthropic's platforms and through the major cloud providers, and developers can build it into their own apps.

GPT-6 Astra is an AI model from OpenAI, the company behind ChatGPT, that can operate a computer to get work done. OpenAI says it can fill out online forms, organize a calendar, research a topic and draft a summary, build a website, and produce documents, spreadsheets and presentations. OpenAI also says it is better than its earlier models at making the right call when instructions are unclear, and that it asks focused questions when the answer could change the outcome. OpenAI said that within days of its launch it would reach all ChatGPT Plus, Pro, Business and Enterprise users and developers building their own apps.

GPT-6.1 Sol is an AI model from OpenAI, an upgrade to GPT-6 Sol made for coding, operating a computer and professional work. OpenAI says it does much better than GPT-6 Sol on complex jobs such as writing and fixing code, understanding documents and carrying out business tasks with many steps. OpenAI, which makes ChatGPT, offers it to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and developers can build it into their own apps.

Claude Sonnet 5.5 is a chat and coding model from Anthropic, an AI safety and research company, and a faster, lower-cost partner to Claude Opus 5.5. Anthropic says it is especially good at well-defined everyday tasks, fixing bugs and making polished documents, slides and spreadsheets, and calls it a clear upgrade over Claude Sonnet 5. People can use it in the Claude apps and Claude Code, and developers can build it into their own apps, including through Amazon, Google and Microsoft cloud services.

GPT-6.1 Terra is a new AI model from OpenAI for the everyday work people do on a computer. OpenAI says it is fast and efficient, pairing a close reading of images with precise use of a computer and clearer, more natural writing, for workday tasks like drafting proposals, reading technical drawings and filling out forms. OpenAI calls it a significant step up from GPT-5.6 Terra. It is out now for Plus, Pro, Enterprise and Edu users in ChatGPT Work and in Codex, and developers can build it into their own apps.

DeepSeek V4.1 Flash 1017 is an update to DeepSeek's V4.1 Flash AI model. DeepSeek says the update is better at reasoning, coding, multistep tasks and tasks with images, and that it finishes tasks faster with shorter answers, which lowers their cost. DeepSeek, an AI company in China, has made it the model behind its chat website and app, and its weights are also free to download.

MiMo-V2.6-Pro is an AI model from Xiaomi that takes in text, images, video and audio, and can keep very large code projects and long work sessions in view. Xiaomi trained it for coding, visual work, cybersecurity and tasks it carries out on its own. Xiaomi, which makes phones, electric cars and other smart devices, has made the model free to download and run. People can also use it in Xiaomi's MiMo desktop app and AI Studio, and developers can build it into their own apps.

Step 5 Preview is an AI model from StepFun built to carry out tasks on its own, from writing software to researching a topic and producing a structured report, and StepFun says it is particularly strong at finance work. It can read long documents, charts, screenshots and images. StepFun, which also makes a family of voice and music models, offers it in its own products and to developers to build into their own apps. Since October 15 the model has also been free to download and run.

GLM-5.3 Flash is a coding model from Z.AI that can take in images, video and files along with text. Z.AI says it watches the screens and pages it builds, and how they respond, to test and improve its own work as it goes. Z.AI traces its start to ZhipuAI, a company that grew out of research at Tsinghua University. The model is free to download and run.

Muse Spark Open is an AI model from Meta, whose products billions of people use, and a version of its Muse Spark that anyone can download. Meta says it understands more than text and is built for coding and long tasks with many steps. Meta also says it is the first model of its kind that can run on a desktop computer, using two Apple Mac Studio computers with M5 Ultra chips. Developers can also use it through several cloud services.

GPT Image 2.5 is an image generator and editor from OpenAI, the company behind ChatGPT. OpenAI says it gives more natural lighting and richer textures, makes images faster than the previous version, and is better at keeping the people and things in reference photos recognizable and at changing only what it is asked to change while leaving the rest of a picture alone. It is in ChatGPT for all users on desktop, mobile and web, and developers can build it into their own apps in two versions, one quicker and one more precise.

Qwen-Image-2.1 is an image model from Alibaba that makes pictures from text descriptions and edits existing ones, all in a single model. It can create and edit images with see-through backgrounds, and it can blend up to 10 reference pictures into one scene, such as six separate portraits brought together in one group photo. It belongs to Qwen, the family of AI models that Alibaba Cloud shares with the open-source community. It is free to download and run under Qwen's research license.

MAI Image 2.6 is an image generator from Microsoft that makes pictures from a text description and edits existing images. Microsoft says it can make precise, targeted changes, such as altering one object, updating text or cleaning up motion blur, while keeping the picture consistent from one edit to the next. It comes after MAI Image 2.5 in Microsoft's own line of image models, which began with MAI Image 1. People can try it in Microsoft's MAI Playground, and developers can build it into their own apps through Microsoft Foundry, where it is in preview.

MOSS-Image is a pair of image generators from OpenMOSS, a research team formed by the Shanghai Innovation Institution, the language lab at Fudan University and the company MOSI Intelligence. One is a base model and the other a version further trained on carefully chosen examples. OpenMOSS says the family draws clear text inside pictures, handles anime and photo-like styles, and lets people set where each object and line of text sits in the frame. Anyone can download both models.

FLUX 3 Image Klein is an image generator and editor from Black Forest Labs, which has labs in Freiburg and San Francisco and says its founders helped pioneer Stable Diffusion. Black Forest Labs says it shares the main abilities of the full FLUX 3 Image model, with pictures up to 4K, precise editing and boxes that set where each element goes, and that it can work from up to ten reference images. The company says it was tuned to run on the graphics cards in ordinary home computers. Anyone can download it, and it is free for research, noncommercial use and smaller companies.

Wan 3.0 is a video generation model from Alibaba that turns text, images, audio, video or documents such as slides, PDFs and spreadsheets into videos up to 30 seconds long. It creates speech, sound effects and background music along with the picture, and Alibaba says it renders lifelike people and stays faithful to reference images. It is the first model in Alibaba's Wan video family that can read documents, and it can change a finished video's visuals, plot and dialogue without starting over. Developers can build it into their own apps through Alibaba Cloud Model Studio.

MiniMax H3 Max is a video generator that turns a written prompt or a still image into a clip of 5 to 15 seconds with matching sound. It is a version of MiniMax's H3 model that fal trained further for closer prompt following, better looks and better sound. According to fal, a five-second clip at 768p resolution can be ready in under three seconds on its own systems. People can make videos with it on MiniMax's Hailuo AI site and on fal, which offers a few free clips a day and lets developers build it into their own apps.

Seedance 2.5 is a video generator from ByteDance that makes clips with sound, built for its products such as Dreamina. It makes clips up to 30 seconds long in one pass, twice the length of the previous version, and ByteDance says it can follow up to 30 images, 10 video clips and 10 audio clips at once as references. People can also change a chosen moment of a video by its timestamp. It is available on Dreamina to subscribers in Europe, Asia, the Middle East and South America.

FLUX 3 Max is a video generator that turns a written prompt or a still image into a short clip. It is a version of the FLUX 3 model from Black Forest Labs that fal's research team trained further with Black Forest Labs, and fal says a ten-second clip comes back in about twelve seconds. Black Forest Labs, which has labs in Freiburg and San Francisco, says its founders helped pioneer Stable Diffusion. Developers can use FLUX 3 Max through fal and build it into their own apps.

Kling 4.0 is the newest Kling video generation model from Kuaishou, a large content and social platform company that also made the earlier Kling 3.0. Kuaishou says it brings big gains in realism, creative control and storytelling, along with better sound, lip sync and text that appears inside the video. It also makes two-channel stereo sound, and Kuaishou says its lip sync now keeps voices and lip movements in step. Kling 4.0 became available to everyone on October 15.

Lyria 3.5 is a music generator from Google that makes full-length songs, with verses, choruses and bridges, from a text description or a picture. Google says it brings significant advances in musicality, lyrics and vocal quality, and gives more control over a track's tempo and length. Google first released it in its Flow Music app and has since added it to the Gemini app. Developers can also build it into their own apps.

Mureka V9.5 is a music generator that turns a genre, a mood or a line of lyrics into a complete song with vocals. Mureka says this version sounds more natural, with better singing, a clearer sense of genre and closer attention to what was asked for. Mureka, a music service run from offices in Singapore, lets people make songs on its website just by typing in lyrics, with no music knowledge needed.

Suno V6 is a family of music models that turn ideas into songs, made by Suno, a company founded in 2022 whose team is mostly musicians. Suno says it can start from a written idea, a voice memo, a picture or a video, and can rework one section of a song from a plain request while keeping everything else. The main v6 model and a less predictable v6-wild are for paying users, while a faster v6-mini is open to everyone.

Eleven Music v2.5 is a music generator from ElevenLabs, a company that started out building lifelike voices and has since moved well beyond them, and it is now the default in its ElevenMusic app. ElevenLabs says its songs sound more layered, with more natural-sounding instruments. It also works in ElevenCreative, and developers can build it into their own apps.

MiniMax Music 3.0 is a music generator from MiniMax that composes, arranges, performs and produces a complete song from a description of the idea, with or without lyrics. MiniMax says it sticks to what the creator asked for across songs up to five minutes long, with clear, realistic instruments and vocals that sound performed rather than synthesized. MiniMax is the AI company behind the Hailuo video models. Music 3.0 is free to download and run.