Archived front page, Wed 30 Sep 2026, 09:00 UTC. All of 2026-09-30 · Archive · Live page

The Daily Markdown

The front page of the internet for AI users · Wed 30 Sep 2026, 09:00 UTC

Top now

6 stories

What is climbing fastest across every source

OpenAI5.8xAnthropic6.5xGLM-5.37.0xLanguage6.6xVideo Generation6.9xGPT-6.17.0xMuse7.4xClaude7.3xRobots6.5xDevDay7.1x

ROSS: Relearning from Self-Generated Rollouts through Selective Supervision

Researchers introduced ROSS, a method that reuses historical model outputs during training by applying loss only to selected continuations. This approach improved the Qwen3.6-35B-A3B model's average benchmark score from 58.40% to 62.20%. ·

HF Daily Papers+ arXiv9h ago▲3 to 16 across 6 pollsarxiv.org

openai/codex 0.159.2

OpenAI released version 0.159.2 of Codex, a patch that stops console windows from flashing on Windows when the tool runs background processes. ·

GitHub Releases9h ago▲0 to 6 across 4 pollsgithub.com

ROSS: Relearning from Self-Generated Rollouts through Selective Supervision

Researchers introduced ROSS, a method that reuses historical model outputs during training by applying loss only to selected continuations. This approach improved the Qwen3.6-35B-A3B model's average benchmark score from 58.40% to 62.20%. ·

HF Daily Papers+ arXiv9h ago▲3 to 16 across 6 pollsarxiv.org

feder-cr/dots

Feder-cr released an open-source AI agent that operates its own browser to avoid being blocked on the web. ·

GitHub10h ago▲133 to 613 across 5 pollsgithub.com

openai/codex 0.159.2

OpenAI released version 0.159.2 of Codex, a patch that stops console windows from flashing on Windows when the tool runs background processes. ·

GitHub Releases9h ago▲0 to 6 across 4 pollsgithub.com

PostHog/jeeves

235 points and 90 comments on Hacker News, linking github.com.

GitHub+ Hacker News23h ago▲116 to 314 across 19 pollsgithub.com

ROSS: Relearning from Self-Generated Rollouts through Selective Supervision

Researchers introduced ROSS, a method that reuses historical model outputs during training by applying loss only to selected continuations. This approach improved the Qwen3.6-35B-A3B model's average benchmark score from 58.40% to 62.20%. ·

HF Daily Papers+ arXiv9h ago▲3 to 16 across 6 pollsarxiv.org

JoasASantos/Offensive-Security-AI-Models

JoasASantos maintains a repository of AI models that are uncensored or fine-tuned for cybersecurity tasks. ·

GitHub+ Hacker News3d ago▲110 to 438 across 42 pollsgithub.com

Trending topics today: heat

OpenAI: 3.14OpenAI3.14Anthropic: 2.30Anthropic2.30GLM-5.3: 1.32GLM-5.31.32Language: 1.23Language1.23Video Generation: 1.19Video Generation1.19GPT-6.1: 0.98GPT-6.10.98
Table
OpenAI3.14
Anthropic2.3
GLM-5.31.32
Language1.23
Video Generation1.19
GPT-6.10.98

Heat = summed hotness of today's stories naming the topic.

Models

5 stories

Releases, papers, trending weights and new listings

Top 6 trending on Hugging Face: likes

convaiinnovations/laya: 4,554convaiinnovations/laya4,554Qwen/Qwen-Image-2.1 ▼1: 2,679Qwen/Qwen-Image-2.1 ▼12,679abenzerps/Qwen-Image-2.1-Uncensored-GGUF: 2,473abenzerps/Qwen-Image-…2,473Edge0/Audio8-ASR-Infinite: 1,559Edge0/Audio8-ASR-Infi…1,559XingChen-AGI/TeleOCR ▲1: 925XingChen-AGI/TeleO… ▲1925Contrastive-LM/CLM-v0.1-8B ▲1: 540Contrastive-LM/CLM… ▲1540
Table
convaiinnovations/laya4554
Qwen/Qwen-Image-2.1 ▼12679
abenzerps/Qwen-Image-2.1-Uncensored-GGUF2473
Edge0/Audio8-ASR-Infinite1559
XingChen-AGI/TeleOCR ▲1925
Contrastive-LM/CLM-v0.1-8B ▲1540

PirateBench: score out of 100

opus-5-5: 91opus-5-591gpt-6-sol: 82gpt-6-sol82deepseek-v4-flash-0731: 63deepseek-v4-flash-073163
Table
opus-5-591.15
gpt-6-sol81.78
deepseek-v4-flash-073162.82

Agents build our own projects from a brief; every run, failures included: https://cowerx.dev/daily/piratebench/

MaLiang-Harness: A Programmable Path to Image and Video Generation

Researchers introduced MaLiang-Harness, a framework that uses multimodal models to iteratively build, check, and fix code for image and video generation. The system evaluates eleven closed-source models on new benchmarks to measure success and quality. ·

HF Daily Papers9h ago▲5 to 105 across 5 pollsarxiv.org

Omni-IO Skills: Harnessing Your Agent Omni-Native

Researchers present Omni-IO Skills, a harness that enables agents to process seven modalities. It includes 27 skills covering 38 tasks and boosts input support rates to 100%. ·

HF Daily Papers9h ago▲15 to 46 across 3 pollsarxiv.org

Language models for text classification: From bag-of-words to Jev

The source text lists titles for visual guides and experiments on recurrent neural networks, convolutional neural networks, and transformers, focusing on their accuracy and efficiency. ·

105 points and 4 comments on Hacker News, linking magazine.sebastianraschka.com.

Hacker News+ Ahead of AI22h ago▲9 to 109 across 39 pollsdiscussmagazine.sebastianraschka.com

Claude Sonnet 5.5

191 points and 138 comments on Hacker News, linking anthropic.com.

Hacker News+ Simon Willison39h ago▲65 to 329 across 4 pollsdiscussanthropic.com

JoasASantos/Offensive-Security-AI-Models

JoasASantos maintains a repository of AI models that are uncensored or fine-tuned for cybersecurity tasks. ·

GitHub+ Hacker News3d ago▲110 to 438 across 42 pollsgithub.com

MaLiang-Harness: A Programmable Path to Image and Video Generation

Researchers introduced MaLiang-Harness, a framework that uses multimodal models to iteratively build, check, and fix code for image and video generation. The system evaluates eleven closed-source models on new benchmarks to measure success and quality. ·

HF Daily Papers9h ago▲5 to 105 across 5 pollsarxiv.org

Omni-IO Skills: Harnessing Your Agent Omni-Native

Researchers present Omni-IO Skills, a harness that enables agents to process seven modalities. It includes 27 skills covering 38 tasks and boosts input support rates to 100%. ·

HF Daily Papers9h ago▲15 to 46 across 3 pollsarxiv.org

X voices

5 stories

X posts other sources are linking to, then named voices on Bluesky and Mastodon

GitHub

5 stories

New repos, releases and changelog

New AI repos this week: stars

KKKKhazix/AIHOT: 3,691KKKKhazix/AIHOT3,691dzhng/jevgrep: 1,825dzhng/jevgrep1,825feitangyuan/onetake: 1,008feitangyuan/onetake1,008kaankiziltug/logo-design-skill: 924kaankiziltug/logo-des…924lemomo-ai/lemo-opuscar: 640lemomo-ai/lemo-opuscar640feder-cr/dots: 613feder-cr/dots613
Table
KKKKhazix/AIHOT3691
dzhng/jevgrep1825
feitangyuan/onetake1008
kaankiziltug/logo-design-skill924
lemomo-ai/lemo-opuscar640
feder-cr/dots613

feder-cr/dots

Feder-cr released an open-source AI agent that operates its own browser to avoid being blocked on the web. ·

GitHub10h ago▲133 to 613 across 5 pollsgithub.com

openai/codex 0.159.1

OpenAI updated Codex version 0.159.1 to make GPT-6.1 Sol the default model in its bundled and Amazon Bedrock catalogs. ·

GitHub Releases12h ago▲0 to 7 across 6 pollsgithub.com

PostHog/jeeves

235 points and 90 comments on Hacker News, linking github.com.

GitHub+ Hacker News23h ago▲116 to 314 across 19 pollsgithub.com

JoasASantos/Offensive-Security-AI-Models

JoasASantos maintains a repository of AI models that are uncensored or fine-tuned for cybersecurity tasks. ·

GitHub+ Hacker News3d ago▲110 to 438 across 42 pollsgithub.com

feder-cr/dots

Feder-cr released an open-source AI agent that operates its own browser to avoid being blocked on the web. ·

GitHub10h ago▲133 to 613 across 5 pollsgithub.com

openai/codex 0.159.1

OpenAI updated Codex version 0.159.1 to make GPT-6.1 Sol the default model in its bundled and Amazon Bedrock catalogs. ·

GitHub Releases12h ago▲0 to 7 across 6 pollsgithub.com

Infrastructure

5 stories

Data centers, power, chips and what tokens cost

Data-center and energy stories per day

Thu 24: 3Thu 243Fri 25: 20Fri 2520Sat 26: 1Sat 261Sun 27: 1Sun 271Mon 28: 20Mon 2820Tue 29: 33Tue 2933Wed 30: 2Wed 302
Table
Thu 243
Fri 2520
Sat 261
Sun 271
Mon 2820
Tue 2933
Wed 302

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

Last $5.00Min $2.85Max $5.00

One point per OpenRouter reading, last 8 days. Data by OpenRouter, CC BY 4.0.

Output price, $ per million tokens (newest models)

gpt-6.1-sol-pro: 10gpt-6.1-sol-pro10gpt-6.1-sol-pro:batch: 5gpt-6.1-sol-pro:batch5gpt-6.1-sol: 10gpt-6.1-sol10gpt-6.1-sol:batch: 5gpt-6.1-sol:batch5claude-sonnet-5.5: 10claude-sonnet-5.510median, all paid models: 2median, all paid mode…2
Table
gpt-6.1-sol-pro10
gpt-6.1-sol-pro:batch5
gpt-6.1-sol10
gpt-6.1-sol:batch5
claude-sonnet-5.510
median, all paid models2

438 paid models listed; median of the 60 added in 30 days: $5.00. Rankings data by OpenRouter, CC BY 4.0.