Encrypted Reasoning Traces on Your Laptop: Resume or Strip
Public traces hid 315,320 encrypted blocks. Treat resume logs and publish logs as two paths, then count the field on your own machine.
20 stories
Public traces hid 315,320 encrypted blocks. Treat resume logs and publish logs as two paths, then count the field on your own machine.

A new open-source Rust tool renders Claude Code sessions as a live, scrubbable flow graph inside your terminal or browser.

Apodex open-sourced FrontierAgent, an Apache 2.0 terminal agent framework with ReAct and Agent Team modes, alongside a 35B open-weight model.

Perplexity's new Portable Computer runs the full agent stack locally on NVIDIA DGX Spark, with cloud escalation gated by user approval and zero per-token cost for on-device work.

Kyutai released the full training stack for its 100M-parameter Pocket TTS, letting anyone train a voice model from scratch for under $200.

Z.ai's new 753B open-weights model keeps the GLM-5.2 base but doubles down on post-training, topping open-source coding and cyber benchmarks.
A 36B Qwen3-MoE with refusals ablated gets a full ladder of GGUF quants, putting an uncensored sparse model on 16GB GPUs.

NVIDIA's new inference accelerator hit 3,431 tokens/second on Gemma 4 31B at 100K context, roughly 4x the fastest public endpoint in third-party testing.

A style LoRA for MiniMax H3 generates video with hand-painted gouache backgrounds and golden-age character animation from text alone.

Google Research drops a 330M-parameter foundation model that forecasts multiple related time series jointly, topping three zero-shot benchmarks but locked to non-commercial use.

Anthropic rebuilt the Claude web and desktop streaming renderer to only update changing text, cutting stalls by 9x and holding 120fps on ProMotion Macs.

The open-source browser agent slots into xAI's coding CLI as a plugin, letting Grok drive a real Chrome session to test sites and squash bugs.

Thinking Machines Lab is offering up to $50,000 in Tinker credits to researchers tackling the hardest open-weight model safety problems, from tamper-resistant safeguards to worst-case risk forecasting.

OpenAI's Sol, Terra, and Luna models land in AWS's agentic IDE, with Terra cutting successful-task costs by roughly 82% on Terminal-Bench 2.1.

Mistral is trading solo AI ambitions for Gulf compute, joining Saudi's HUMAIN in a hundreds-of-millions-of-euros push into sovereign, Arabic-first models.
A community re-quantization of Ornith-1.5-35B-A3B revives its multi-token prediction head, matching Opus 4.6 on SWE-bench-Live at 23 GB.

Pipecat released PhoneLLM Alpha 1, a 30B Mamba-Transformer MoE fine-tuned for voice phone agents with sub-100ms TTFT and $0.00025 per agent-minute.

Anthropic ships an open MCP extension that pipes enterprise SSO straight into Claude connectors, killing the per-user OAuth consent tax.
Mia-AiLab shipped a 5.0bpw EXL3 quant of the DFlash2 speculative drafter for Qwen3.8-27B, cutting draft weight from 3.85 GB to 1.4 GB and lifting decode throughput 33 percent on DGX Spark.

Prime Intellect's open-source harness turns agents into programmable systems, driving ARC-AGI-3 scores from 30% to 95.5% and enabling week-long autonomous runs.