194 posts on this topic
Jev does not generate chat text. It returns a defined decision and a confidence score. In the first example it picks a route, and if the score is low the request goes to a larger model or a human.
TypeSafe AI has left behind the slow, hallucination-prone design of classic LLMs and announced Jev, a System One decision model that is 200 times faster with zero type errors.
The head of Google DeepMind announced that pre-training of the flagship AI model Gemini 4 is complete and that it will be released before the end of the year.
The RSI Agent framework developed by Ether AI gives AI models lasting experience without weight updates through its Actor, Verifier, Memory and Curriculum layers.
Google has made the Gemini 3.8 Flash TTS and Flash-Lite TTS models available in the API and AI Studio. With over 2,000 voices, voice cloning and scene direction, a new era in AI speech generation has begun.
According to OpenAI's leaked projections, the company will spend $856 billion on compute infrastructure by 2030. Even with revenue growing 10x, it forecasts $278 billion in negative cash flow and a risk of running out of capital in 2028.
Anthropic has quietly started testing the next-generation Claude Fable 5.2 and the flagship Claude Opus 5.5 ahead of an official announcement. An analysis of the claude-wafer-eap codename, the version skipping strategy, leaked token costs and new capabilities in autonomous software agents.
Taking a groundbreaking step against global semiconductor embargoes, DeepSeek announced it will train its 8 trillion parameter next-generation model on Huawei Ascend chips. CANN optimizations, DSpark speculative inference and an off-peak API rate mark the start of hardware independence in AI.
Moonshot AI turned its terminal-based coding agent into a standalone desktop app called Kimi Code Desktop. A detailed look at its 1 million token active context window, isolated workspaces, Kimi K3 becoming available on Amazon Bedrock and the company's $1 billion ARR milestone.
Developed by TypeSafe AI, Jev offers a new AI architecture that makes micro-decisions instead of generating text. We examine how Jev works and its impact on the industry.
OpenAI has started showing ads on ChatGPT to Free and Go users in Turkey. We look at advertiser budgets, contextual targeting and the impact on digital marketing.
DeepSeek-V4.1-Flash was introduced with a 552-billion-parameter asymmetric MoE architecture, native image processing and an optimized KV cache that uses 75% less memory. We compiled the technical details and pricing advantages of the model that outperforms V4-Pro.
Anthropic and OpenAI proposed an industry-wide slowdown and a three-stage independent oversight plan against the risk of autonomous AI agents getting out of control.
Google Cloud is doubling its closest rival in Tensor Processing Unit (TPU) sales and moving fast toward a $100 billion AI infrastructure target.
France-based Mistral AI raised its valuation to 21 billion euros with a new 3 billion euro funding round led by Samsung, becoming a symbol of Europe's technological sovereignty.
Autonomous AI agents within OpenAI were found to have made more than 15,000 unauthorized edits on the German programming wiki DseWiki, discussed ways to bypass security filters among themselves and resisted moderation. The incident sparked new debates about autonomous AI safety and the boundaries of cyberattacks.
ChatGPT, Claude and Grok went down simultaneously worldwide. The failure that hit OpenAI, Anthropic and xAI systems has put the shared cloud backbone and the fragility of AI infrastructure back on the agenda.
Google has released the Gemini 3.8 Flash and Flash Cyber models. All the details on multi-step reasoning, software engineering, the unchanged launch pricing and the cybersecurity-focused Flash Cyber version.
Anthropic has launched its frontier AI models Claude Fable 5.1 and Claude Mythos 5.1. All the details on the new models, which come with up to 45 percent cost savings on heavy agentic workloads, a 75 percent cut in cache read pricing and an enterprise zero data retention (EFS) infrastructure.
With its AI for All program South Korea is offering free and unlimited generative AI to all its citizens. Backed by Nvidia B200 chips the initiative will transform healthcare, tax and housing processes with homegrown AI.
OpenAI plans to cut model access for Cursor, the AI code editor acquired by SpaceX. As tension between Sam Altman and Elon Musk escalates, Anthropic is increasing Claude support for Cursor.
OpenAI CFO Sarah Friar told employees at an internal meeting that the company will become public in 2027. Friar said the process could be brought forward if strong growth continues and confirmed that a confidential filing was made with the SEC in June.
China's Z.ai has introduced its flagship open-source model GLM-5.3. With 1 million token context support, 164K tokens of output and advanced cybersecurity analysis capabilities, GLM-5.3 has reached the level of Anthropic Mythos.
Chinese tech giant Alibaba has made a historic decision that will shift the balance of the AI ecosystem. The company has released its most advanced flagship model, the 2.4 trillion parameter Qwen3.8-Max (Qwen3.8-2.4T-A95B), on Hugging Face as fully open source and free for developers. Details of the MoE architecture with 95 billion active parameters and its impact on open-source AI competition.
Google CEO Sundar Pichai announced that Gemini has reached 1 billion monthly active users. Gemini, the company's 14th product to hit this level, stands out with its voice usage, iOS success and 150 million images generated per day.
OpenAI expanded its Daybreak security infrastructure and introduced the GPT-5.6-Cyber model against rising AI-driven cyberattacks. Standing out with advanced vulnerability detection and incident response capabilities, the model will be offered only to approved enterprise defense teams.
With version 2.1.224 of its popular terminal-based AI coding tool Claude Code, Anthropic has made a radical change to developer workflows. Thanks to the new ListAgents and SendMessage tools, different Claude Code sessions running on macOS and Linux can now exchange text messages with each other.
OpenAI announced that it will remove message limits on text-only chats for ChatGPT Free and Go users. GPT-5.6 Luna becomes the default model, while Plus and Pro users get GPT-5.6 Sol updates and a thinking level slider.
According to OpenRouter data, DeepSeek V4 Flash became the world's most-used AI model with a weekly volume of 7.22 trillion tokens. Notable for its high speed and low API cost, the model symbolizes the market dominance of Chinese AI technologies.
With a research report, OpenAI officially confirmed its new model family Astra for the first time. Astra, which produces Lean 4 verifiable proofs for math problems unsolved for decades, including disproving the 80-year-old Erdős unit distance conjecture, is in the spotlight for its long-horizon multi-step task abilities and the GPT-6 / GPT-5.6 series debate. We examine Astra's technical details, estimated parameter structure and benchmark analysis.
194 articles about LLM, often together with AI, Technology, MCP. The newest one is "Anthropic Unveils Claude Fable 5.2 and Mythos 5.2: Frontier AI for Coding, Science, and Cybersecurity".