ai-model

European startup Domyn announces a new 400 billion parameter AI model, following their existing 260b closed-source model and a 10b open-source model.

Qwen released a new large language model, Qwen-AgentWorld-397B-A17B, as detailed on HuggingFace and the Qwen blog.

Mistral OCR 4 converts handwritten calculus exams into clean LaTeX, accurately reading formulas and accounting for graphs, though it does not redraw them. The model provides structured output with bounding boxes and confidence scores in 170 languages.

A diffusion model that can transform any image into an interactive, playable hallucination, running locally on user hardware.

Asks for recommendations on affordable AI models for content writing, image generation, and vibe coding.

A promotional video showcasing the capabilities of Seedance 2.5, an AI video generation model.

The author praises GLM-5.2, an MIT open-weights model, for its exceptional real-world performance in human evaluation benchmarks, claiming it rivals the best closed-source models like those from Claude.

Muennighoff announces work on a much better composer model, scaling to Opus/GPT-size, training from scratch, and going beyond coding, as part of Cursor's collaboration with SpaceX.

Sakana AI announces Fugu Ultra, a multi-agent orchestration model that matches frontier performance of Fable and Mythos while avoiding export controls.

An updated GPT-5.5 Cyber model surpasses Mythos 5 in the CyberGym benchmark.

A tweet reacts to a claim that AEON-7 is the best model distill, leaving even the creator speechless.

OpenAI releases the full version of GPT-5.5-Cyber, a cybersecurity-focused AI model with state-of-the-art performance on CyberGym, and announces efforts to improve security through Patch The Planet and Codex Security.

According to leaked information, OpenAI's GPT‑5.6 is nearly complete, expected to have a 1.5 million context window, priced at only one-third of Fable 5, with significantly improved front-end generation quality. Release may be delayed until after the 28th.

This tweet points out that you can deploy the GLM-5.2 model locally for 128,000 yuan.

The Gemma 4 QAT 31B model demonstrates improved behavior with KV cache quantization, suggesting enhanced inference efficiency.

Video of Michael Truell announcing Cursor's new AI model at the Compile event.

Sakana AI's new model Fugu outperforms Fable 5 on LiveCodeBench and Terminal Bench 2.1 by small margins, according to their corporate blog, though results are not yet independently confirmed.

A new Japanese AI model achieves performance comparable to leading American frontier models, marking a significant advancement.

Sakana in Japan released a competitor to the Mythos model, and initial impressions suggest it looks great.

A user expresses confusion about the status or behavior of the Claude Opus 4.8 AI model, prompting discussion.