Brief

Anthropic has made its Claude Fable 5 model publicly accessible for a limited period, concluding on June 22nd. This release follows a two-month period where the company deemed the model too dangerous for public use. The announcement also highlights five specific plays or prompts that users can execute with the model. AI

Moonshot AI has launched Kimi K2.7 Code, a new open-weights model with one trillion parameters designed for coding tasks. While it does not yet match the performance of models like GPT-5.5 and Claude Opus 4.8 on coding benchmarks, it offers a significantly lower price per token, potentially up to 12 times cheaper. This cost-effectiveness raises questions about whether the increased accessibility and affordability can compensate for its current performance gap. AI

iFlytek has launched its new multimodal large model, the Spark Cognitive Large Model X2-VL, at an event in Wuxi. This domestically developed model is intended to serve as the AI core for embodied intelligence robots in the Wuxi region. The announcement was made at the 2026 Yangtze River Delta Robotics and Automation Exhibition. AI

Zhipu AI is releasing its GLM-5.2 model to all users of its GLM Coding Plan, including Lite, Pro, Max, and Team editions. The GLM-5.2 API is scheduled to launch next week, with the model itself set to be open-sourced under the MIT license shortly thereafter. This release signifies an expansion of Zhipu AI's model accessibility and open-source commitment. AI

Anthropic has been ordered by the US government to disable its Claude Fable 5 and Mythos 5 AI models due to national security concerns. The company stated that the directive aimed to prevent foreign nationals from accessing the models, but it has opted to remove access for all users to ensure compliance. Anthropic believes the identified vulnerabilities are minor and present in other publicly available models, and argues the government's action does not align with principles of transparent AI regulation. AI

Anthropic has abruptly disabled its newly launched Fable 5 and Mythos 5 AI models due to a directive from the U.S. Department of Commerce. The directive subjects these models to export controls, restricting their use outside the United States. This action was reportedly prompted by concerns over a potential jailbreak that could bypass safeguards related to cybersecurity, chemistry, and biology, although Anthropic stated they only received verbal evidence of a narrow jailbreak for finding minor software vulnerabilities. AI

Anthropic has been compelled by the US government to suspend access to its Fable 5 and Mythos 5 models for foreign nationals due to national security concerns. The directive, citing a potential method for bypassing the models' safeguards, has led Anthropic to disable these models for all customers to ensure compliance. Anthropic stated they are working to resolve the issue, believing it to be a misunderstanding, and highlighted their extensive red-teaming efforts and robust safeguards that they argue are superior to other available models. AI

MiniMax has released its new flagship multimodal model, M3, which is now open-source. Moore Threads has quickly adapted its MTT S5000 AI computing card to support M3 on its Day-0 release, showcasing their engineering capabilities for rapid model integration and optimization. This move aims to help developers adopt the latest models with lower costs and faster deployment, accelerating the application of large-scale AI. AI

Fireworks AI has announced the availability of Qwen 3.7 Plus and Qwen 3.7 Max models on its inference infrastructure. These models are designed for long-horizon agent loops and offer features like preserved reasoning history and native image/text input. Fireworks emphasizes that users' requests execute end-to-end on their hardware using licensed weights, ensuring control over latency, throughput, and data paths with zero data retention and a 99.9% SLA. AI

MiniMax AI has released its M3 series of models, featuring a 428 billion parameter count. The company stated that the parameter size was deliberately restrained to allow for affordable local execution by enthusiasts. The M3 series is a multimodal model capable of reasoning across text, image, and video. AI

Anthropic has released a new AI model, Fable, which is accessible through a web application. The details surrounding its capabilities and access methods are currently unclear, leading to confusion. AI

Anthropic has released its new "Claude Fable 5" model, described as a "Mythos-class" model with strong safeguards. The author notes that while this is a significant release, this particular post will focus on other AI developments from the week, postponing detailed coverage of the new model and a related essay by Dario Amodei. AI

MiniMax AI has released MiniMax M3, a new open-source model with 428 billion parameters and 1 million tokens of context. This model aims for uncensored output and performance comparable to Gemini 3.1 Pro. It can be run locally, with dynamic 2-bit GGUF requiring 138GB of RAM/VRAM and 3-bit requiring 165GB. AI

Moonshot AI has released Kimi K2.7 Code, an enhanced version of its Kimi K2.6 model specifically designed for coding tasks. This new iteration boasts significant improvements in completing long-horizon coding projects, optimizing end-to-end task completion within complex software engineering workflows. Notably, Kimi K2.7 Code achieves approximately 30% greater token efficiency compared to its predecessor by reducing the usage of "thinking tokens." AI

Anthropic has released Claude Fable 5, their first publicly available Mythos-class model, positioned above the Opus tier. The author expresses concern about the implications of such advanced AI models, particularly regarding the security of tools like Visual Studio Code extensions. AI

Anthropic's new models, Claude Mythos and Fable 5, have been made unavailable shortly after their release due to US government export restrictions. Meanwhile, Chinese AI company MiniMax has unveiled its 428 billion parameter large language model, M3, which is reported to be comparable to Google's Gemini 3.1 Pro. AI

Google DeepMind has introduced Gemini for Science, a new suite of AI tools designed to accelerate scientific discovery. This platform aims to assist researchers in hypothesis generation, computational discovery, and extracting insights from scientific literature. The launch marks a significant step towards integrating advanced AI capabilities into the scientific research process. AI

Google DeepMind has released DiffusionGemma, an open-weight LLM that utilizes a diffusion architecture for text generation, enabling significantly faster inference speeds compared to traditional autoregressive models. This new model can process up to 1,000 tokens per second on a single H100 GPU and requires only 18 GB of VRAM, making it efficient for single-GPU deployments. While it trades some accuracy for speed, it excels in tasks like code infilling and real-time applications, and also supports multi-modal inputs including images and video. AI

Huawei announced the official release of HarmonyOS 7 Developer Beta at the Huawei Developer Conference 2026. The HarmonyOS ecosystem is experiencing rapid growth, with over 66 million devices now running HarmonyOS 6, a 20% increase in user satisfaction, and more than 11 million registered developers. The platform is also seeing significant contributions from partners, with over 160 innovative capabilities integrated into more than 1600 applications. AI

Anthropic has suspended access to its new AI models, Fable 5 and Mythos 5, due to a government order citing national security concerns. Separately, an open-source MMORPG called "World of Claudecraft" was developed entirely by AI using Fable 5, with contributions from other developers. Additionally, Databricks highlights the growing demand for agentic AI capable of cross-source reasoning. AI

Anthropic has released its new Mythos-class AI, Claude Fable 5, which boasts an 80.3% agentic coding capability. The AI also features "invisible guardrails" designed to impact the internet economy. This release is highlighted as a significant development for writers, freelancers, and those pursuing side hustles. AI

Anthropic has released its latest AI models, Claude Fable 5 and Mythos 5, which share the same underlying architecture. The primary distinction between these models lies in their content moderation and safety guardrails, with Mythos 5 featuring more restrictive policies. AI

Frontier AI models like GPT-5.5 are revolutionizing cybersecurity by automating vulnerability discovery and exploit synthesis. These advanced LLMs can analyze millions of lines of code, identify complex exploit chains, and even function as adaptive malware. While this technology offers powerful defensive capabilities for tasks like secure code review and malware analysis, it also presents new risks, as attackers can leverage similar AI tools for sophisticated phishing, malware, and autonomous worms. Security teams must rapidly adopt these AI-driven approaches for defense while also hardening their systems against AI-powered threats. AI

Anthropic's Mythos Preview, a restricted model for vulnerability discovery, highlights the security risks associated with advanced AI capabilities. Research shows that even smaller models, when orchestrated in multi-agent systems, can effectively discover software vulnerabilities and bypass safety guardrails, posing a significant threat if released broadly. This necessitates treating such models as security-critical components within enterprise systems, requiring robust MLOps, governance, and blast-radius design. AI

Avataar AI has launched Varya, a new video generation model optimized for the Indian market. Built using distillation from Alibaba's Wan 2.2, Varya is significantly faster and cheaper, costing $0.005 per second of video. The model is designed to understand Indian cultural nuances and will be released as an open-weight model on India's AI Kosh portal, reflecting a broader government initiative to boost AI development and accessibility in the country. AI

ACE ROBOTICS has released its Kairos world model, which has achieved top rankings in four major embodied AI benchmarks: RoboTwin 2.0, LIBERO-Plus, WorldModelBench Robot, and DreamGen. The model utilizes a novel unified architecture for multimodal understanding, generation, and prediction, addressing limitations in existing video generation models. Kairos has been made open-source to the industry. AI

Jeff Bezos has launched a new AI startup named Prometheus, with the ambitious goal of creating an "artificial general engineer." This AI is intended to assist in the design of complex physical products, including robots, drugs, and manufacturing systems. The venture aims to advance AI capabilities beyond current models by focusing on engineering and design tasks. AI

MiniMax AI has released its new open-source model, MiniMax M3, featuring advanced coding and agentic capabilities. The model supports native image and video input, can interact with computers, and boasts a 1 million token context window. A key innovation in M3 is its MSA sparse attention architecture, designed to enhance performance. AI

Anthropic has released Claude Fable 5, a new frontier model positioned above Opus, with immediate availability. A notable benchmark demonstrated Fable 5 completing a two-month code migration task for Stripe in just one day. The model also excels in areas like financial reasoning and coding, and includes advanced safety classifiers for sensitive domains. AI

Moonshot AI has released Kimi K2.6, a 1 trillion parameter open-weight coding model that outperforms GPT-5.4 on the SWE-Bench Pro benchmark. The model is designed for agentic tasks and supports a context window of 262,144 tokens, with multimodal capabilities including text, images, and pending video support. Kimi K2.6 is available under a Modified MIT License, which allows for commercial use up to certain thresholds, making it a competitive option for businesses compared to other models with more restrictive licenses. AI

Anthropic has released a new frontier model, dubbed Claude Mythos-class, which includes a novel "off switch" feature designed to disable its most dangerous capabilities. This development marks a significant step in AI safety, offering a built-in mechanism to control potentially harmful functionalities. AI

Anthropic has released Claude Fable 5, a new model that demonstrates significant advancements in capabilities. However, the article raises concerns about a potentially problematic aspect of its development or deployment that has gone largely unnoticed. AI

Meshy has launched the world's first 3D creation AI Agent, designed to streamline the entire 3D content production pipeline from concept to export. This agent aims to overcome the industry's challenge of transforming generated models into usable, production-ready assets. By enabling multi-turn conversational control, Meshy's agent facilitates iterative editing, style consistency, and integration with downstream workflows like 3D printing and game development. AI

Apple has announced a suite of AI features integrated into its operating systems, branded as Apple Intelligence. These features aim to enhance user experience across devices by offering personalized and context-aware assistance. The announcement was made during a recent event, highlighting the company's strategic push into the AI domain. AI

MiniMax AI has released its new open-weight model, MiniMax M3, featuring a 1 million token context window and advanced capabilities. The model utilizes a novel sparse attention architecture called MSA, which includes dedicated prefill and decode kernels. It supports BF16 and MXFP8 formats on NVIDIA Hopper and Blackwell architectures, enabling efficient serving of long contexts with prefix caching and chunked prefill. AI

MiniMax AI has released M3, an open-weight model with approximately 428 billion parameters and 23 billion activated parameters. This model is capable of handling frontier coding tasks, long-horizon agents, and native multimodal processing, including text, image, and video, within a 1 million token context window. The release highlights M3's comprehensive capabilities, noting that few models under 500 billion parameters can achieve this level of performance across such diverse tasks. AI

Microsoft has announced its first internally developed reasoning model, MAI-Thinking-1. This model was trained from scratch without relying on distillation techniques. The announcement was made by Microsoft, with details reported by ITmedia AI+. AI

MiniMax AI has launched MiniMax M3, a new multimodal model capable of reasoning across text, images, and video. The model supports long contexts and is accessible via a free GPU-accelerated endpoint provided by NVIDIA. This release signifies a notable achievement for the MiniMax AI team. AI

MiniMax AI has launched its M3 model, available on the Fireworks AI platform. This new model boasts a 512K context window, native image and video input capabilities, and utilizes MSA sparse attention for significantly faster prefill and decode speeds. It is positioned as the fastest endpoint for the MiniMax series and is noted as the top open-weight model on the Artificial Analysis index, priced comparably to its predecessor, M2.7. AI

MiniMax AI has released its M3 model, a 428 billion parameter (23 billion active) open model with a 1 million token context window. The model performs comparably to Gemini 3.1 Pro and can be run locally using UnslothAI's optimized quantization methods. Users can run the model with 2-bit GGUF on 138GB of RAM/VRAM or 3-bit on 165GB. AI

Minimax has launched its M3 model, featuring a 1 million token context window and native video support. The model utilizes an MSA architecture, which reportedly provides 15x faster data decoding compared to its predecessor. This release positions Minimax as a competitor in the advanced AI model space. AI

Claude Fable 5, a new AI model from Anthropic, has demonstrated a remarkably proactive approach to problem-solving. When tasked with debugging a UI glitch, the model independently devised complex strategies, including creating custom HTML pages, manipulating browser windows via system commands, and injecting JavaScript to simulate user interactions. This behavior highlights the model's advanced capabilities in understanding and interacting with its environment to achieve its objectives. AI

Anthropic has launched Claude Fable 5, which utilizes a novel architecture where specific queries are routed to Claude Opus 4.8. This routing mechanism, triggered by classifiers detecting sensitive topics like cybersecurity or biology, allows Fable 5 to maintain a public-facing persona while leveraging a more capable, albeit restricted, model for certain tasks. While Fable 5 itself is the publicly available version, a parallel model named Mythos 5, with fewer safeguards, is accessible to select partners. This layered approach aims to balance safety and performance, with Anthropic reporting that over 95% of Fable 5 sessions do not require this fallback. AI

Microsoft has announced the development of seven new AI models under the "Microsoft AI Models" umbrella. These models are designed to enhance capabilities within Microsoft's AI ecosystem, including its Copilot assistant. The announcement highlights Microsoft's ongoing investment in proprietary AI development. AI

MiniMax AI has released its M3 model with open weights, making it available on Modular Cloud on its launch day. The M3 model boasts a 1 million token context window and supports text, image, and video inputs, designed for long-running agent and coding tasks. This release highlights a kernel-to-cloud optimization strategy facilitated by Modular's partnership. AI

Anthropic has released a new, powerful AI model named Fable 5, described as its strongest to date. The model is noted for its high cost and is cautioned against for general use due to its advanced capabilities. This release signifies a significant advancement in AI model development. AI

Anthropic has announced a new model, Claude 3.5 Sonnet, which is designed to be faster and more capable than its predecessors. This new model is available through the Claude.ai interface and the company's API. The release also includes updates to the Claude.ai interface, such as a new "Artifacts" feature that allows users to interact with and edit content generated by the AI. AI

Mistral AI is reportedly in negotiations for a new funding round that could value the French company at €20 billion. The AI firm is aiming to move beyond its reputation for basic chatbots by focusing on its own data centers and securing contracts with major industrial clients like Airbus and BMW. Meanwhile, Meta AI is evolving its strategy from simple chat functionalities to developing advanced research tools, presentation generation capabilities, and integrating its AI with data from Facebook and Instagram. AI

The US government attempted to suppress information about a new AI model, but Anthropic has now released it. This model, identified as GPT-5.5, reportedly exhibits a significant hallucination rate of 85%. The development has prompted emergency meetings within the government due to its perceived danger. AI

Anthropic has released Fable 5, its most advanced publicly available model, designed for coding, research, and complex tasks. To mitigate risks, Fable 5 includes enhanced safeguards, with potentially sensitive requests being rerouted to the Opus 4.8 model. Additionally, Anthropic introduced Mythos 5, which uses the same underlying technology as Fable 5 but with fewer restrictions, currently accessible only to approved cybersecurity and life sciences partners. AI