Archived front page, Tue 20 Oct 2026, 01:00 UTC. All of 2026-10-20 · Archive · Live page

The Daily Markdown

The front page of the internet for AI users · Tue 20 Oct 2026, 01:00 UTC · Get it by email

Top now

6 stories

What is climbing fastest across every source

LLM7.3xDeepSeek7.8xMercury 37.1xAnthropic5.1xOpenRouter6.8xLearning7.5xFLUX-36.8xLong-Horizon Agents5.1xVideo World4.8xWorld Modeling4.4x

Token-maxxing is dead, OpenRouter's token usage plateaus

OpenRouter reports that token usage has plateaued, indicating that the trend of maximizing token consumption has ended. The platform tracks real-world usage by the number of tokens processed through its API. ·

27 points and 25 comments on Hacker News, linking openrouter.ai.

Hacker News1h ago▲11 to 52 across 4 pollsdiscussopenrouter.ai

anthropics/claude-code v2.1.304

Anthropic releases Claude Code version 2.1.304, which adds new mod hooks, a guided setup command, and a security comparison tool. ·

GitHub Releases3h ago▲6 to 13 across 2 pollsgithub.com

My obituary for programming

Jeremy Scheff writes that coding agents cause career programmers to lose the satisfaction of solving logic puzzles, while those who prefer building systems find large language models beneficial. ·

Simon Willison1h ago▲simonwillison.net

Token-maxxing is dead, OpenRouter's token usage plateaus

OpenRouter reports that token usage has plateaued, indicating that the trend of maximizing token consumption has ended. The platform tracks real-world usage by the number of tokens processed through its API. ·

27 points and 25 comments on Hacker News, linking openrouter.ai.

Hacker News1h ago▲11 to 52 across 4 pollsdiscussopenrouter.ai

anthropics/claude-code v2.1.304

Anthropic releases Claude Code version 2.1.304, which adds new mod hooks, a guided setup command, and a security comparison tool. ·

GitHub Releases3h ago▲6 to 13 across 2 pollsgithub.com

vercel-labs/fx0

Vercel Labs releases fx0, a lightweight design agent built with libfx that generates a single HTML file from a user's page description. ·

GitHub8h ago▲github.com

unslothai/unsloth DeepSeek-V4.1 & FLUX-3

Unsloth releases DeepSeek-V4.1-Flash-1017 and FLUX.3 Image Klein for local use, adding image and video chat features, multi-GPU training, and one-command MCP setup for coding agents. ·

GitHub Releases9h ago▲11 to 14 across 4 pollsgithub.com

finereli/whatsapp-mcp

A read-only MCP server lets AI agents search and read personal WhatsApp messages while running on Cloudflare Workers. ·

A new repository written in TypeScript with 285 stars in its first week.

GitHub+ Hacker News28h ago▲111 to 285 across 20 pollsgithub.com

anthropics/claude-code v2.1.304

Anthropic releases Claude Code version 2.1.304, which adds new mod hooks, a guided setup command, and a security comparison tool. ·

GitHub Releases3h ago▲6 to 13 across 2 pollsgithub.com

Token-maxxing is dead, OpenRouter's token usage plateaus

OpenRouter reports that token usage has plateaued, indicating that the trend of maximizing token consumption has ended. The platform tracks real-world usage by the number of tokens processed through its API. ·

Hacker News1h ago▲11 to 52 across 4 pollsdiscussopenrouter.ai

deepseek-ai/DeepSeek-V4.1-Flash-1017

A image text to text model from deepseek-ai with 1,693 likes and 17.9k downloads on Hugging Face.

Hugging Face+ Hacker News3d ago▲226 to 1693 across 128 pollshuggingface.co

vercel-labs/fx0

Vercel Labs releases fx0, a lightweight design agent built with libfx that generates a single HTML file from a user's page description. ·

GitHub8h ago▲github.com

Trending topics today: heat

LLM: 2.05LLM2.05DeepSeek: 1.62DeepSeek1.62Mercury 3: 1.45Mercury 31.45Anthropic: 1.35Anthropic1.35OpenRouter: 1.04OpenRouter1.04Learning: 0.75Learning0.75
Table
LLM2.05
DeepSeek1.62
Mercury 31.45
Anthropic1.35
OpenRouter1.04
Learning0.75

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

google/embeddinggemma-2 ▼2: 2,725google/embeddingge… ▼22,725jialinyyzz/humanizer ▼2: 1,987jialinyyzz/humaniz… ▼21,987deepseek-ai/DeepSeek-V4.1-Flash-1017: 1,693deepseek-ai/DeepSeek-…1,693openbmb/MiniCPM-V-4.7-35B-A3B: 1,066openbmb/MiniCPM-V-4.7…1,066black-forest-labs/flux-3-image-klein-comfy ▲3: 964black-forest-labs/… ▲3964fishaudio/drama-3 ▲1: 770fishaudio/drama-3 ▲1770
Table
google/embeddinggemma-2 ▼22725
jialinyyzz/humanizer ▼21987
deepseek-ai/DeepSeek-V4.1-Flash-10171693
openbmb/MiniCPM-V-4.7-35B-A3B1066
black-forest-labs/flux-3-image-klein-comfy ▲3964
fishaudio/drama-3 ▲1770

PirateBench: score out of 100

gpt-6.1-sol: 99gpt-6.1-sol99opus-5-5: 96opus-5-596gpt-6-sol: 92gpt-6-sol92gpt-6-luna: 65gpt-6-luna65haiku-4-5: 4.28haiku-4-54.28
Table
gpt-6.1-sol98.9
opus-5-596.46
gpt-6-sol91.6
gpt-6-luna65.25
haiku-4-54.28

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

unslothai/unsloth DeepSeek-V4.1 & FLUX-3

Unsloth releases DeepSeek-V4.1-Flash-1017 and FLUX.3 Image Klein for local use, adding image and video chat features, multi-GPU training, and one-command MCP setup for coding agents. ·

New release v0.1.909-beta of unslothai/unsloth on GitHub.

GitHub Releases9h ago▲11 to 14 across 4 pollsgithub.com

Towards Company World Models

The text describes AI as a strategic partner for essential businesses, where it learns from owner-operators to propose, implement, and measure better decisions. ·

Hacker News6h ago▲3 to 20 across 9 pollsdiscussbeaconholdings.com

deepseek-ai/DeepSeek-V4.1-Flash-1017

A image text to text model from deepseek-ai with 1,693 likes and 17.9k downloads on Hugging Face.

Hugging Face+ Hacker News3d ago▲226 to 1693 across 128 pollshuggingface.co

unslothai/unsloth DeepSeek-V4.1 & FLUX-3

Unsloth releases DeepSeek-V4.1-Flash-1017 and FLUX.3 Image Klein for local use, adding image and video chat features, multi-GPU training, and one-command MCP setup for coding agents. ·

GitHub Releases9h ago▲11 to 14 across 4 pollsgithub.com

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

feder-cr/git-ai: 2,453feder-cr/git-ai2,453Leonxlnx/earshot: 1,536Leonxlnx/earshot1,536aicodeking/leanrea: 993aicodeking/leanrea993OpenMOSS/MOSS-Image: 949OpenMOSS/MOSS-Image949LiCinCan/blender-ai-mcp-addon: 931LiCinCan/blender-ai-m…931alchaincyf/huashu-motion: 917alchaincyf/huashu-mot…917
Table
feder-cr/git-ai2453
Leonxlnx/earshot1536
aicodeking/leanrea993
OpenMOSS/MOSS-Image949
LiCinCan/blender-ai-mcp-addon931
alchaincyf/huashu-motion917

vercel-labs/fx0

Vercel Labs releases fx0, a lightweight design agent built with libfx that generates a single HTML file from a user's page description. ·

A new repository written in TypeScript with 115 stars in its first week.

GitHub8h ago▲github.com

unslothai/unsloth DeepSeek-V4.1 & FLUX-3

Unsloth releases DeepSeek-V4.1-Flash-1017 and FLUX.3 Image Klein for local use, adding image and video chat features, multi-GPU training, and one-command MCP setup for coding agents. ·

GitHub Releases9h ago▲11 to 14 across 4 pollsgithub.com

finereli/whatsapp-mcp

A read-only MCP server lets AI agents search and read personal WhatsApp messages while running on Cloudflare Workers. ·

GitHub+ Hacker News28h ago▲111 to 285 across 20 pollsgithub.com

finereli/whatsapp-mcp

A read-only MCP server lets AI agents search and read personal WhatsApp messages while running on Cloudflare Workers. ·

A new repository written in TypeScript with 285 stars in its first week.

GitHub+ Hacker News28h ago▲111 to 285 across 20 pollsgithub.com

vercel-labs/fx0

Vercel Labs releases fx0, a lightweight design agent built with libfx that generates a single HTML file from a user's page description. ·

GitHub8h ago▲github.com

unslothai/unsloth DeepSeek-V4.1 & FLUX-3

Unsloth releases DeepSeek-V4.1-Flash-1017 and FLUX.3 Image Klein for local use, adding image and video chat features, multi-GPU training, and one-command MCP setup for coding agents. ·

GitHub Releases9h ago▲11 to 14 across 4 pollsgithub.com

Infrastructure

5 stories

Data centers, power, chips and what tokens cost

Data-center and energy stories per day

Wed 14: 34Wed 1434Thu 15: 41Thu 1541Fri 16: 35Fri 1635Sat 17: 5Sat 175Sun 18: 2Sun 182Mon 19: 30Mon 1930Tue 20: 0Tue 200
Table
Wed 1434
Thu 1541
Fri 1635
Sat 175
Sun 182
Mon 1930
Tue 200

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 $3.20Min $3.20Max $5.00

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

Output price, $ per million tokens (newest models)

deepseek-v4.1-flash-1017: 0.60deepseek-v4.1-flash-1…0.60glimpse: 3glimpse3beam-1.0-501b-a23b: 1beam-1.0-501b-a23b1mercury-3: 0.35mercury-30.35moondream3-preview: 2.50moondream3-preview2.50median, all paid models: 2median, all paid mode…2
Table
deepseek-v4.1-flash-10170.6
glimpse3
beam-1.0-501b-a23b1
mercury-30.35
moondream3-preview2.5
median, all paid models2

441 paid models listed; median of the 50 added in 30 days: $3.20. Rankings data by OpenRouter, CC BY 4.0.

Detecting model weight tampering in GPU memory (Rome attacks)

Stealthium detects ROME attacks, where malicious Python modules rewrite model weights in GPU memory, by linking CUDA kernel launches to their originating code. ·

15 points and 1 comments on Hacker News, linking stealthium.io.

Hacker News10h ago▲5 to 16 across 13 pollsdiscussstealthium.io

Belief-Trajectory Energy: Measuring the Path to a Prediction

Researchers introduce Belief-Trajectory Energy, a metric that tracks how large language models revise predictions across layers. The method achieves 0.998 macro-AUROC for detecting human-written text. ·

HF Daily Papers25h ago▲2 to 16 across 21 pollsarxiv.org

Detecting model weight tampering in GPU memory (Rome attacks)

Stealthium detects ROME attacks, where malicious Python modules rewrite model weights in GPU memory, by linking CUDA kernel launches to their originating code. ·

15 points and 1 comments on Hacker News, linking stealthium.io.

Hacker News10h ago▲5 to 16 across 13 pollsdiscussstealthium.io

Belief-Trajectory Energy: Measuring the Path to a Prediction

Researchers introduce Belief-Trajectory Energy, a metric that tracks how large language models revise predictions across layers. The method achieves 0.998 macro-AUROC for detecting human-written text. ·

HF Daily Papers25h ago▲2 to 16 across 21 pollsarxiv.org