Woman arrested after blocking delivery robot in Center City Philadelphia
39 points and 27 comments on Hacker News, linking 6abc.com.
Top now
6 storiesWhat is climbing fastest across every source
Anthropic6.0xblack-forest-labs6.8xflux-36.8xChinese7.1xScion7.7xLLMs5.1xLearning5.4xOpenRouter5.8xAnalyzing6.8xMononoke7.0x
[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment
Nick Cammarata claims that agents disprove polysemanticity, greatly improving interpretability and alignment. ·
MAI-Gym: An Open All-in-One Platform for GUI Agent Evaluation and RL Training across Mobile and Computer Environments
MAI-Gym is an open platform for evaluating and training GUI agents across mobile and computer environments. It supports 20,000 parallel instances and trains a 9B agent. ·
Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation
Researchers introduce Waypoint-OPD, a distillation method that gates policy shifts using intermediate checkpoints to improve transfer between Qwen3 model sizes in reasoning tasks. ·
PointWormGazebo/panel-all-in-one-any-game
PointWormGazebo offers a universal cheat panel that provides AI-enhanced aimbot and per-game precision tuning for various video games. ·
Woman arrested after blocking delivery robot in Center City Philadelphia
39 points and 27 comments on Hacker News, linking 6abc.com.
Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation
Researchers introduce Waypoint-OPD, a distillation method that gates policy shifts using intermediate checkpoints to improve transfer between Qwen3 model sizes in reasoning tasks. ·
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
Researchers introduce OmniCompass, a benchmark for evaluating multimodal models on video grounding and understanding. It includes 820 vision-language videos and 240 audio-visual videos. ·
PointWormGazebo/panel-all-in-one-any-game
PointWormGazebo offers a universal cheat panel that provides AI-enhanced aimbot and per-game precision tuning for various video games. ·
Call Center Jobs Are Declining, but 14,000 Calls Show Humans Still Outperform AI
New research from Bold Agent Kit indicates that human receptionists achieve roughly four times higher inbound intake conversion rates than AI, despite AI handling a larger volume of calls. ·
openai/codex 0.163.0
OpenAI releases Codex version 0.163.0, which adds parallel read-only tool execution and enables Daybreak by default for eligible accounts. ·
Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation
Researchers introduce Waypoint-OPD, a distillation method that gates policy shifts using intermediate checkpoints to improve transfer between Qwen3 model sizes in reasoning tasks. ·
125 upvotes on Hugging Face Daily Papers.
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
Researchers introduce OmniCompass, a benchmark for evaluating multimodal models on video grounding and understanding. It includes 820 vision-language videos and 240 audio-visual videos. ·
Woman arrested after blocking delivery robot in Center City Philadelphia
39 points and 27 comments on Hacker News, linking 6abc.com.
Call Center Jobs Are Declining, but 14,000 Calls Show Humans Still Outperform AI
New research from Bold Agent Kit indicates that human receptionists achieve roughly four times higher inbound intake conversion rates than AI, despite AI handling a larger volume of calls. ·
anthropics/claude-code v2.1.301
New release v2.1.301 of anthropics/claude-code on GitHub.
openai/codex 0.163.0
OpenAI releases Codex version 0.163.0, which adds parallel read-only tool execution and enables Daybreak by default for eligible accounts. ·
Trending topics today: heat
Table
| Anthropic | 2.2 |
| black-forest-labs | 1.07 |
| flux-3 | 1.07 |
| Chinese | 0.95 |
| Scion | 0.83 |
| LLMs | 0.82 |
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
| abenzerps/Qwen-Image-2.1-Uncensored-Quantized | 4402 |
| google/embeddinggemma-2 | 2438 |
| jialinyyzz/humanizer | 1770 |
| Qwen/Qwen-Image-2.1-Turbo | 900 |
| openbmb/MiniCPM-V-4.7-35B-A3B | 806 |
| fishaudio/drama-3 ▲4 | 499 |
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/
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
Researchers introduce OmniCompass, a benchmark for evaluating multimodal models on video grounding and understanding. It includes 820 vision-language videos and 240 audio-visual videos. ·
113 upvotes on Hugging Face Daily Papers.
Self-Evolving Agent Harness
Harness self-evolution revises agent software scaffolding while keeping model weights fixed. The framework separates update targets from proposal and admission mechanisms. ·
OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
OpenGCUA introduces an open stack for generalist computer-use agents, improving task scores on OSWorld-v2 across eight models, including GPT-6 Astra and Claude Opus 5. ·
RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment
Researchers introduce RoTA-Bench, a benchmark containing 674 tasks that evaluates vision-language models on 3D spatial reasoning through interactive projection alignment. ·
Learning Where to Split: Dynamic Band Splitting for Audio Source Separation
Researchers introduce Dynamic Band Splitting, an audio separation framework that adapts frequency boundaries to specific mixtures. The method achieves a 29.0% relative gain in average SI-SDR on the DnR v2 dataset. ·
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
Researchers introduce OmniCompass, a benchmark for evaluating multimodal models on video grounding and understanding. It includes 820 vision-language videos and 240 audio-visual videos. ·
113 upvotes on Hugging Face Daily Papers.
Self-Evolving Agent Harness
Harness self-evolution revises agent software scaffolding while keeping model weights fixed. The framework separates update targets from proposal and admission mechanisms. ·
OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
OpenGCUA introduces an open stack for generalist computer-use agents, improving task scores on OSWorld-v2 across eight models, including GPT-6 Astra and Claude Opus 5. ·
RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment
Researchers introduce RoTA-Bench, a benchmark containing 674 tasks that evaluates vision-language models on 3D spatial reasoning through interactive projection alignment. ·
Learning Where to Split: Dynamic Band Splitting for Audio Source Separation
Researchers introduce Dynamic Band Splitting, an audio separation framework that adapts frequency boundaries to specific mixtures. The method achieves a 29.0% relative gain in average SI-SDR on the DnR v2 dataset. ·
X voices
5 storiesX posts other sources are linking to, then named voices on Bluesky and Mastodon
@OpenAI
GPT-Live-1.1 – New voices. Better accents. Less "checking."
2 points and 0 comments on Hacker News, linking an X post.
@SEGA
Total War Remaster Removed from Steam After AI Allegations, Brings Back Original
4 points and 0 comments on Hacker News, linking an X post.
@opencode
DeepSeek v4.1 Flash is now free in OpenCode
4 points and 0 comments on Hacker News, linking an X post.
@firesidealpha
AI researchers are taking up smoking to signal to other researchers
3 points and 0 comments on Hacker News, linking an X post.
@HelloSurgeAI
Can Coding Agents Act Like Staff Engineers?
2 points and 1 comments on Hacker News, linking an X post.
@OpenAI
GPT-Live-1.1 – New voices. Better accents. Less "checking."
2 points and 0 comments on Hacker News, linking an X post.
@SEGA
Total War Remaster Removed from Steam After AI Allegations, Brings Back Original
4 points and 0 comments on Hacker News, linking an X post.
@opencode
DeepSeek v4.1 Flash is now free in OpenCode
4 points and 0 comments on Hacker News, linking an X post.
@OpenAINewsroom
OpenAI finds recording device in one of their offices
85 points and 2 comments on Hacker News, linking an X post.
@HelloSurgeAI
Can Coding Agents Act Like Staff Engineers?
2 points and 1 comments on Hacker News, linking an X post.
GitHub
5 storiesNew repos, releases and changelog
New AI repos this week: stars
Table
| onehumanbeing/esp32-iphone-agent | 2670 |
| rociiu/talorys | 1210 |
| deeplethe/solid-glass | 1174 |
| marclou/mailcheap | 1085 |
| zhuyansen/new-term-radar | 1038 |
| feder-cr/open_scion | 1023 |
openai/codex 0.163.0
OpenAI releases Codex version 0.163.0, which adds parallel read-only tool execution and enables Daybreak by default for eligible accounts. ·
New release rust-v0.163.0 of openai/codex on GitHub.
anthropics/claude-code v2.1.301
New release v2.1.301 of anthropics/claude-code on GitHub.
ggml-org/llama.cpp v0.7.0
llama.cpp v0.7.0 adds LiliCorr speculative decoding, a GPU cache for host-memory experts, and support for models like K2 Horizon and Prism Bonsai 2. ·
unslothai/unsloth NVIDIA Nemotron 3.5 + Meta Muse Spark Open
Unsloth adds local support for NVIDIA Nemotron 3.5 and Meta Muse Spark Open, introducing text-diffusion LLMs, DeepSpeed training, and improved prompt caching for coding agents. ·
feder-cr/open_scion
Feder-cr releases open_scion, an open-source version control system designed specifically for coding agents. ·
anthropics/claude-code v2.1.301
New release v2.1.301 of anthropics/claude-code on GitHub.
openai/codex 0.163.0
OpenAI releases Codex version 0.163.0, which adds parallel read-only tool execution and enables Daybreak by default for eligible accounts. ·
ggml-org/llama.cpp v0.7.0
llama.cpp v0.7.0 adds LiliCorr speculative decoding, a GPU cache for host-memory experts, and support for models like K2 Horizon and Prism Bonsai 2. ·
feder-cr/open_scion
Feder-cr releases open_scion, an open-source version control system designed specifically for coding agents. ·
unslothai/unsloth NVIDIA Nemotron 3.5 + Meta Muse Spark Open
Unsloth adds local support for NVIDIA Nemotron 3.5 and Meta Muse Spark Open, introducing text-diffusion LLMs, DeepSpeed training, and improved prompt caching for coding agents. ·
Infrastructure
5 storiesData centers, power, chips and what tokens cost
Data-center and energy stories per day
Table
| Sat 10 | 3 |
| Sun 11 | 7 |
| Mon 12 | 41 |
| Tue 13 | 29 |
| Wed 14 | 34 |
| Thu 15 | 41 |
| Fri 16 | 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
| moondream3-preview | 2.5 |
| moondream3.1-9b-a2b | 2.5 |
| muse-spark-open | 4.25 |
| nemotron-3.5-super-vl | 0.5 |
| pareto-latest | 7.5 |
| median, all paid models | 2 |
438 paid models listed; median of the 47 added in 30 days: $5.00. Rankings data by OpenRouter, CC BY 4.0.
Market selloff, rising rates test AI data centre deal boom
49 points and 40 comments on Hacker News, linking reuters.com.
Reliability Standards for AI Data Centers, Crypto Miners Move Closer to Adoption
Proposed reliability standards for large AI data centers and crypto miners await approval by NERC and federal regulators, applying to loads of 50 MW or more at 100 kV or higher. ·
Banks are now trying to offload the AI data center debt they just underwrote
378 points and 274 comments on Hacker News, linking startupfortune.com.
Market selloff, rising rates test AI data centre deal boom
49 points and 40 comments on Hacker News, linking reuters.com.
Banks are now trying to offload the AI data center debt they just underwrote
378 points and 274 comments on Hacker News, linking startupfortune.com.
Nvidia's first GPU: GeForce 256
60 points and 39 comments on Hacker News, linking dfarq.homeip.net.
AI models may soon self-replicate without access to data centers
25 points and 39 comments on Hacker News, linking reuters.com.
AI models may be capable of self-replicating without access to a datacenter
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
