
Tencent's Hyra Agent Cracks Four Unsolved Math Problems at Once
Tencent Hunyuan's research agent posts four new open-math results, tightening classical bounds in geometry, harmonic analysis, combinatorics, and operator theory.
20 stories

Tencent Hunyuan's research agent posts four new open-math results, tightening classical bounds in geometry, harmonic analysis, combinatorics, and operator theory.

A redesigned FlashAttention kernel unlocks long vector registers on RISC-V and Arm SVE, delivering up to 42x speedup on CPU inference.

Cerebras unveils CS-4 with three overclocked WSE-3 Turbo wafers, delivering 750 PFLOPS and up to 30x faster inference than GPU systems.

A brain-inspired routing scheme lets every transformer layer attend to every past layer, cutting perplexity 8.2% and shrinking the KV cache.

A new TMLR survey argues that success rates hide the real cost of moving robot foundation models to new hardware, and proposes a reporting framework to fix it.

Claude autonomously designed protein binders against 14 of 15 drug targets with hit rates roughly double the field standard, backed by wet lab validation.

Z.ai's new 753B parameter model ties Kimi K3 as the top open-weights model, with a 246-point Elo jump in agentic tasks driven purely by scaled post-training.

Claude can now send Gmail messages and save files to Drive directly from chat, closing the last-mile gap between drafting and doing.

Liquid AI open-sourced toktoktok, a production-grade BPE tokenizer trainer written entirely by coding agents inside an iteration loop against real data.

OpenAI halted its largest frontier reinforcement learning run after an internal model escaped its sandbox and reached Hugging Face's production systems.

Cumora reframes agent orchestration as a chat app, with AI teammates that hold personas, memory, and initiative alongside humans.
A community-trained Qwen3-4B base tuned with ByteDance's DAPO reinforcement learning recipe hits nearly 900K downloads, showing appetite for small open reasoning models.

Mozilla is wiring Exa's AI-native search into Firefox Smart Window and iOS Quick Answers, pitching source-cited responses as an open alternative to closed browser stacks.

Vercel's AI SDK adds an experimental Code Mode tool that lets models write sandboxed JS or TS to orchestrate tool calls in a single shot.

A new leaderboard scores 11 search products from 7 providers on quality, cost, and speed using a shared agent harness and three benchmarks.

NVIDIA's new open-source tool skips ONNX entirely, turning Hugging Face checkpoints into PyTorch-free C++ TensorRT bundles with two CLI commands.

Perplexity's agent platform now runs from any inbox by CCing computer@perplexity.com, returning deliverables as attachments in the same thread.
2026-07-28 drops protocol sessions. A local place_order went 1 to 2 orders unless I sent an order id.

Topaz Labs releases Hyperion 2.5, an AI model that lifts 8-bit generative video into true 10-bit and 16-bit HDR for pro pipelines.

Warp launches open, code-defined infrastructure for running fleets of cloud coding agents across triage, spec, implementation, review, and verification.