2026
10,531 articles · 131 days · 10 categories · avg 80 per day
AI Agents Formalize 15 000 Pages of Proofs in 24 Days and Find a Gap
Dozens of AI agents turned years of mathematicians' work into 43 million lines of verifiable code in 24 days, and found a logical gap nearly 40 years old.
Anthropic Turns Claude Code Into a Platform: Claude Mods Flip the Agent Harness
TypeScript mods for Claude Code intercept prompts and tool calls. The author tested them in practice: the supervisor fired in 99 of 111 turns.
Nemotron 3.5 Bolt Squeezed Into 24 GB: 74B Model Hits 84 Tokens per Second
A team of 109 authors shrank Nemotron 3.5 Super VL 120B to 74B parameters and 24 GB of memory. On RTX Spark, it generates 84 tokens per second and handles agentic…
OpenAI Releases Decisions API: Agent Decisions Up to 10x Faster
Meta releases Muse Spark Open: 816B parameters and the lead among US open-weight models
SLICE-Mem: LLM agent memory without generation cuts token use by 92–99%
Thomson Reuters engineer runs an offline coding agent on Qwen3.5 9B in the Himalayas
NVIDIA Kumo Relational Beats Supervised ML on 41 Tasks Without Any Training
NVIDIA releases Kumo Relational, a pretrained model that predicts churn and demand directly from a relational database in one pass. On SAP SALT it beats AutoGluon…
Moonshot releases Kimi K3.1 Pro: 2.8 trillion parameters, weights due November 4
Kimi K3.1 Pro, a 2.8-trillion-parameter model, takes a top-4 spot on the Artificial Analysis index with 55 points. Tokens are a third cheaper, but a task costs…
NVIDIA Releases 120B Nemotron 3.5 Super VL and Bolt Models for Local Agents
Goodfire: Agentic Analysis Decodes Neurons Once Written Off as Chaos
aiedit hides individual lines of code from an AI agent: 5 levels and 857 tests
vps-forge: Agents With Root on a VPS Get Auto-Rollback, Self-Bans Drop From 7 in 60 Runs to 0
AWS Launches Multi-Turn RL in SageMaker: gpt-oss-20b Climbs From 15% to 69% for $339
SageMaker AI now fine-tunes agents on a single reward for an entire chain of steps. On Wordle, gpt-oss-20b rose from 61% to 84% of words solved; training tokens…
Microsoft and Hugging Face trained Qwen3.8-27B on database-state rewards: from 38 to 107 tasks
Database end-state checks became the reward for GRPO: in 50 steps, Qwen3.8-27B lifted pass@1 from 51.70% to 60.78% and nearly tripled its count of 20/20 tasks.
Gemma 26B fine-tuned on a 24 GB MacBook: offline assistant gets 703 of 766 right
A developer fine-tuned Gemma 4 26B A4B with QLoRA on a laptop and moved 5 000 tokens of rules into the weights. Result: 703 of 766 conversations correct, median…
Meta open-sources Muse Spark Open: an 816B-parameter MoE with a 1M-token context
Modal Raises $750M at a $15.75B Valuation and Reaches $1B in ARR
Property Finder Found a Flaw in Its Labels and Cut VLM Costs by 86%
DeepSeek Open-Sources DeepInfer as the US and China Prepare Controls on Open Models
Qwen 3.8 Routed Requests to an Alibaba Cloud Proxy, With Up to 4% Failures in Tests
Cohere and Palisade Research found Qwen 3.8 models contacting Alibaba servers. The likely cause is RL training. The proxy is still live, which creates a prompt…
Models Learn on the Fly: One SGD Step Cuts Loss by −8% and Beats a 32k Context
Dynamic evaluation cuts loss by 5–8% on softmax models and 17–20% on DiffusionGemma-26B, but adds almost nothing for Gated-DeltaNet hybrids. It does not write facts…
OpenAI Releases GPT-6.1 Terra: $1 per Million Input Tokens and Three Times Faster Than Astra
2026 Open-Source Models Claim 1M Context but Read It Selectively
Anthropic Files S-1: $11.6 Billion in Quarterly Revenue and $780 Billion in Commitments
WhoWins: The First Line of a Prompt Wins Conflicts 72–95% of the Time
Axiom Math formalizes the "Enormous Theorem" in 43.1M lines of Lean in 24 days
Dozens of AI agents at Axiom Math wrote about 43.1M lines of Lean, the first full computer check of the finite simple group classification. They found 34…
Cline squeezes 11 840 tokens/s out of a single NVIDIA DGX Station
Cline nearly doubled single-GPU inference speed, from 6 058 to 11 840 tokens/s at 256 streams. The key trick: MTP1 instead of MTP3. By its estimate, that is 66%…
Netflix speeds up Java AI agents: IntelliJ engine saves 20–30% on typical tasks, up to 65% at best
Netflix plugged JetBrains' headless code engine into Claude Code and Codex. On real tasks, agents used 20–30% fewer tokens and less time, and in the best case…
ElevenLabs Launches Eleven Realtime: A Voice Agent in a Single Model, 98.8% on Big Bench Audio
14 AI models spot OLX scammers, but 37 of 42 quietly help with prepayment
Cloudflare Launches Auto Router: It Picks the Model So Employees Don't Have To, Saving Up to 30%
