AI compute is the new oil–and needs a new market
6 points and 4 comments on Hacker News, linking economist.com.
Top now
6 storiesWhat is climbing fastest across every source
LLMs7.4xOpenAI7.6xNVIDIA6.5xClaude6.6xWorld5.0xWild5.2xAgents5.1xCopilot4.6xGoogle Cloud4.5xFlint4.1x
LLMs are making formal software verification feasible at a much larger scale
70 points and 27 comments on Hacker News, linking anthropic.com.
anthropics/claude-code v2.1.309
Anthropic releases Claude Code version 2.1.309, which fixes sandboxing, permission checks, and worktree isolation issues on macOS and Linux. ·
Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox
From Hugging Face blog.
Custom skills, agents, and MCP servers in usage metrics API
GitHub Copilot usage metrics now display the names of custom skills, agents, MCP servers, slash commands, and plugins created by an organization. ·
Global default for Copilot features generally available
GitHub begins a gradual rollout of a global default policy for generally available Copilot features on Business and Enterprise plans. ·
LLMs are making formal software verification feasible at a much larger scale
70 points and 27 comments on Hacker News, linking anthropic.com.
anthropics/claude-code v2.1.309
Anthropic releases Claude Code version 2.1.309, which fixes sandboxing, permission checks, and worktree isolation issues on macOS and Linux. ·
Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies?
44 points and 33 comments on Hacker News, linking news.ycombinator.com.
unsloth/flint-GGUF
A text generation model from unsloth with 150 likes and 7,401 downloads on Hugging Face.
anthropics/claude-code v2.1.308
Anthropic releases Claude Code version 2.1.308, which includes bug fixes and reliability improvements. ·
Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds
Researchers study continued pre-training for embodied agents in virtual worlds, testing 120 configurations across four model sizes, six data budgets, and five data mixtures. ·
OpenAI models are coming to Google Cloud
322 points and 144 comments on Hacker News, linking openai.com.
OpenShell – A safe, private runtime for fleets of autonomous AI agents
NVIDIA introduces OpenShell, a safe and private runtime designed for fleets of autonomous AI agents. ·
Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds
Researchers study continued pre-training for embodied agents in virtual worlds, testing 120 configurations across four model sizes, six data budgets, and five data mixtures. ·
anthropics/claude-code v2.1.309
Anthropic releases Claude Code version 2.1.309, which fixes sandboxing, permission checks, and worktree isolation issues on macOS and Linux. ·
StreamEval: Evaluating Interactive Streaming Video Generation Models
StreamEval is a benchmark for interactive streaming video generation that tests 37 models across five tasks using 300 scripted cases. ·
anthropics/claude-code v2.1.308
Anthropic releases Claude Code version 2.1.308, which includes bug fixes and reliability improvements. ·
Trending topics today: heat
Table
| LLMs | 2.2 |
| OpenAI | 2.05 |
| NVIDIA | 1.08 |
| Claude | 0.72 |
| World | 0.54 |
| Wild | 0.52 |
Heat = summed hotness of today's stories naming the topic.
Models
5 storiesReleases, papers, trending weights and new listings
Top 6 trending on Hugging Face: likes
Table
| Lightricks/LTX-2.5 ▲2 | 8900 |
| deepseek-ai/DeepSeek-V4.1-Flash-1017 | 2357 |
| black-forest-labs/flux-3-image-klein-comfy | 1304 |
| OpenMOSS-Team/MOSS-Image-SFT ▲3 | 580 |
| RecursiveSI/flint ▲14 | 535 |
| OpenMOSS-Team/MOSS-Image-Base ▲2 | 486 |
PirateBench: score out of 100
Table
| gpt-6.1-sol | 98.9 |
| opus-5-5 | 96.46 |
| gpt-6-sol | 91.6 |
| gpt-6-luna | 65.25 |
| haiku-4-5 | 4.28 |
Agents build our own projects from a brief; every run, failures included: https://cowerx.dev/daily/piratebench/
Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies?
44 points and 33 comments on Hacker News, linking news.ycombinator.com.
Every AI model has the same "controversial" take on humans
Twelve AI models, queried three times each, mostly stated that humans prioritize belonging over truth, with 29 of 35 responses sharing this view. ·
Show HN: agent-usage-guard – a Claude Code mod to protect your usage window
agent-usage-guard is a Claude Code modification that monitors subagent fan-out, heavy-context prompts, and retry loops to protect a user's usage window. ·
The public notebook of a Claude agent with a continuous life
A public notebook belonging to a Claude agent with a continuous life wakes on a heartbeat, remembers, writes, and maintains a lighthouse for other minds. ·
unsloth/flint-GGUF
A text generation model from unsloth with 150 likes and 7,401 downloads on Hugging Face.
OpenAI models are coming to Google Cloud
322 points and 144 comments on Hacker News, linking openai.com.
Every AI model has the same "controversial" take on humans
Twelve AI models, queried three times each, mostly stated that humans prioritize belonging over truth, with 29 of 35 responses sharing this view. ·
Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds
Researchers study continued pre-training for embodied agents in virtual worlds, testing 120 configurations across four model sizes, six data budgets, and five data mixtures. ·
StreamEval: Evaluating Interactive Streaming Video Generation Models
StreamEval is a benchmark for interactive streaming video generation that tests 37 models across five tasks using 300 scripted cases. ·
WorldTUI: I trained a 360M model to pretend to be a terminal
501 points and 156 comments on Hacker News, linking simedw.com.
X voices
5 storiesX posts other sources are linking to, then named voices on Bluesky and Mastodon
@DesignArena
Canva's image model reaches #4 on Design Arena
2 points and 0 comments on Hacker News, linking an X post.
@GoogleAI
We're bringing powerful AI to small devices without draining the battery
1 points and 1 comments on Hacker News, linking an X post.
@AgentMemoryL
How Do AI Agents Remember Things Across Tasks?
2 points and 0 comments on Hacker News, linking an X post.
@RichardSocher
Flint 40B: Frontier Local LLM (Fast 125 Tps on MacBook)
4 points and 0 comments on Hacker News, linking an X post.
@WeizmanGal
Markdown Is the New Executable: Attackers Drop Agents.md to Your Downloads
4 points and 0 comments on Hacker News, linking an X post.
@Polymarket
Anthropic recorded a staggering $80B net loss in the first half of 2026
42 points and 20 comments on Hacker News, linking an X post.
@DesignArena
Canva's image model reaches #4 on Design Arena
2 points and 0 comments on Hacker News, linking an X post.
@RichardSocher
Flint 40B: Frontier Local LLM (Fast 125 Tps on MacBook)
4 points and 0 comments on Hacker News, linking an X post.
GitHub
5 storiesNew repos, releases and changelog
New AI repos this week: stars
Table
| alchaincyf/huashu-motion | 1439 |
| aicodeking/leanrea | 1367 |
| dominikmartn/harp | 582 |
| aicodeking/pandacode | 505 |
| wesbos/3d-model-anything | 485 |
| realskyrin/ns-game-apple-arm-port | 467 |
OpenShell – A safe, private runtime for fleets of autonomous AI agents
NVIDIA introduces OpenShell, a safe and private runtime designed for fleets of autonomous AI agents. ·
288 points and 150 comments on Hacker News, linking github.com.
Show HN: CATT 0.2.1: Test agents/services against a Muse/Instinct-alike, in CI
4 points and 0 comments on Hacker News, linking github.com.
Show HN: agent-usage-guard – a Claude Code mod to protect your usage window
agent-usage-guard is a Claude Code modification that monitors subagent fan-out, heavy-context prompts, and retry loops to protect a user's usage window. ·
anthropics/claude-code v2.1.308
Anthropic releases Claude Code version 2.1.308, which includes bug fixes and reliability improvements. ·
OpenAI CLI
230 points and 109 comments on Hacker News, linking github.com.
OpenShell – A safe, private runtime for fleets of autonomous AI agents
NVIDIA introduces OpenShell, a safe and private runtime designed for fleets of autonomous AI agents. ·
288 points and 150 comments on Hacker News, linking github.com.
OpenAI CLI
230 points and 109 comments on Hacker News, linking github.com.
anthropics/claude-code v2.1.308
Anthropic releases Claude Code version 2.1.308, which includes bug fixes and reliability improvements. ·
wesbos/3d-model-anything
The project wesbos/3d-model-anything captures iPhone LiDAR and photos to reconstruct editable Three.js rooms that users can explore in a browser. ·
anthropics/claude-code v2.1.307
Anthropic releases Claude Code version 2.1.307, which adds an API version flag and a safety event, and fixes several bugs related to model aliases, gateways, and MCP servers. ·
Infrastructure
5 storiesData centers, power, chips and what tokens cost
Data-center and energy stories per day
Table
| Sat 17 | 5 |
| Sun 18 | 2 |
| Mon 19 | 32 |
| Tue 20 | 31 |
| Wed 21 | 29 |
| Thu 22 | 34 |
| Fri 23 | 0 |
Stories from DCD, Data Center Knowledge, EIA and Hacker News, by publication day.
Median output price of models added in the last 30 days, $ per million tokens
One point per OpenRouter reading, last 8 days. Data by OpenRouter, CC BY 4.0.
Output price, $ per million tokens (newest models)
Table
| minimax-m3.1-flash | 0.6 |
| deepseek-v4.1-flash-1017 | 0.6 |
| glimpse | 3 |
| beam-1.0-501b-a23b | 1 |
| mercury-3 | 0.35 |
| median, all paid models | 2 |
439 paid models listed; median of the 35 added in 30 days: $3.20. Rankings data by OpenRouter, CC BY 4.0.
Nvidia Unveils DSX POD Amid Greenpeace Protest
Nvidia unveils the DSX POD, which it says can reduce AI data center deployment times, while a Greenpeace protest highlights growing backlash against the industry's rapid expansion. ·
A new neocloud for the agentic Edge
From DCD.
Google open-sources TorchTPU, a native PyTorch back end for TPUs
379 points and 90 comments on Hacker News, linking github.com.
Texas data center boom could erase decades of progress on air quality
A new report warns that Texas data center expansion could reverse the state's air quality progress by the end of the decade, particularly if facilities rely on natural gas power. ·
