Tencent Opens Hy4 Flash Weights With 1M-Token Context at 20x Lower Cost
Tencent has released the weights for Hy4 flash, a 220B-parameter multimodal MoE model with 12B active parameters and a 1M-token context, plus FP8 and base versions.
20 stories
Tencent has released the weights for Hy4 flash, a 220B-parameter multimodal MoE model with 12B active parameters and a 1M-token context, plus FP8 and base versions.
Pooling more training corpora can make speech deepfake detectors worse. IDFE adds a corpus classifier with gradient reversal so the model can't lean on dataset quirks, and it cuts pooled error by up to 20%.
GPT-6.2 Astra tops Vals AI's new multi-turn coding benchmark, which checks whether agents can keep extending a working app through up to ten requests without breaking existing features.
The ZazenCodes Season 3 repo is trending because of its Claude Code "software factory" templates. They turn labeled GitHub issues into reviewed pull requests, and a human still does the final merge.
Anthropic adds an opt-in Latest release channel to Claude Desktop, letting users pull fixes and features ahead of the stable track.
Google AI Studio now lets you attach your own branded domain to apps built in Build mode, moving beyond the default ai.studio subdomains.
Unsloth's dynamic GGUF quants shrink Ornith-1.5-122B down to a 40GB file, letting a frontier-class reasoning and coding MoE fit on a single 80GB GPU.
Epoch AI benchmarked Fable 5.2, GPT-6.2 Astra, and four other frontier models on real workplace tasks. None met the employee bar.
Perplexity has opened up the model routing layer it uses in production. One API key now reaches 45+ open and proprietary models, with automatic provider failover, one bill, and an optional router that picks a model for each request.
Microsoft AI drops seven models at Ignite that undercut GPT and Claude on price while matching performance on everyday coding, image editing, voice and security tasks.
GitHub's standalone Copilot app now clones Azure DevOps repos, drives PRs to merge with agents, and adds browser automation plus general availability for GHES.
Microsoft Copilot SDK 2.0 rebuilds the agent runtime in Rust so one process can serve many users. Sessions can now survive restarts, and each one runs with its own identity and tools.
An open-sourced agent skill methodically hill-climbs local LLM serving speed across any hardware and quantization, with measured gains up to 2.1x on prefill.
An open-source browser experiment called Spaceship IDE lets you steer an LLM's next sentences with knobs, in realtime, like piloting a ship.
Frontier coding agents found 'significant' effects in data that contained none in up to 98% of runs. A new replay test corrects the p-values by measuring the agent's own selection bias.
Krea's new Video Swap lets creators replace people, outfits, products, or settings in existing clips while preserving motion, camera moves, and sound.
A three-sensor neural network trained on simulation data reconstructs the full temperature and flow fields of a real natural circulation nuclear test rig.
Anthropic launches linked billing in beta, letting customers approve third-party platforms like Linear, Factory, and Ramp without handing over API keys.
OpenAI drops GPT-6.2 Astra and Sol into ChatGPT Work, Codex, and the API, promising frontier results at a third the cost per task.
Goodfire's unsupervised method finds thousands of tiny neuron groups inside Llama-3.1-8B, often under ten neurons, that are causally necessary and sufficient for behaviors like grammar, arithmetic, and formatting.