Frontier Daily (Aug 25): 15 items from GitHub (Anthropic), Anthropic, GitHub (ModelScope) and more + X supplements

Every day Laojin pulls the official channels and compiles the past 48 hours of AI news into a short brief: curation only, no commentary, with the original link and publish time attached to every item.

1. anthropics/claude-plugins-community: Community plugin marketplace for Claude Cowork and Claude Code. Read-only mirror — submit plugins at clau.de/plugin-directory-submission.

Anthropic has a community plugin marketplace for Claude Cowork and Claude Code, a read-only mirror; submissions go to clau.de/plugin-directory-submission.

Source: GitHub (Anthropic) · 2026-08-24 · Original

2. How Claude's text watermarking works

Future Claude models will include a text watermark to gauge AI authorship, in line with the EU AI Act.

Source: Anthropic News（官网监测） · 2026-08-24 · Original

3. modelscope/FunASR: Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

FunASR is an open-source speech toolkit for ASR, streaming, VAD, punctuation, diarization, and OpenAI-compatible/MCP serving.

Source: GitHub (ModelScope) · 2026-08-24 · Original

4. Your executable is a SQLite database

A Linux trick lets a SQLite database file act as an executable by using the application ID and binfmt_misc.

Source: Simon Willison's Weblog · 2026-08-24 · Original

5. Anthropic’s best AI model struggles to attract users as cheaper tools thrive

Anthropic's July annualized revenue hit $65bn while its best model sees weaker usage compared to cheaper tools, and OpenAI surpassed $40bn.

Source: Simon Willison's Weblog · 2026-08-23 · Original

6. Show HN: Make your logo extra bright on HDR screens

Adding a gain-map to JPEG makes logos brighter on HDR screens, and LinkedIn is one site that preserves it.

Source: Hacker News · 2026-08-22 · Original

7. Why your local LLM feels dumber than it is

A Hacker News post asks why local LLMs seem dumber than they are, drawing 417 points and 171 comments.

Source: Hacker News · 2026-08-22 · Original

8. OpenAI Cuts GPT-5.6 Sol API Pricing by Over 20%

OpenAI drops API and credit pricing for GPT-5.6 Sol by more than 20% for the next 3 months, now available on API and eligible ChatGPT plans while keeping core subscriptions unchanged.

Source: X (Grok x_search, @OpenAI) · within 48h · Original

9. Sam Altman Pauses Frontier RL Training for Safety

Sam Altman announces OpenAI is pausing some frontier RL training to meet alignment, security and monitoring standards as capabilities advance rapidly, stressing that safety will increasingly set the pace of AI progress.

Source: X (Grok x_search, @sama) · within 48h · Original

10. DeepSeek Releases V4-Flash-Vision-Exp Multimodal Model

DeepSeek launches experimental multimodal DeepSeek-V4-Flash-Vision-Exp on its API, matching text performance while making major gains on multimodal agent benchmarks, with support for images via multiple methods and new Files API.

Source: X (Grok x_search, @deepseek_ai) · within 48h · Original

11. Anthropic Uses Claude to Accelerate Protein Binder Design

Anthropic reports Claude achieving 22-35% success rate in designing protein binders versus 10-15% typical, with some designs binding tighter than published ones; technical report and data open-sourced.

Source: X (Grok x_search, @AnthropicAI) · within 48h · Original

12. Google DeepMind Explores Continual Learning in Games

Google DeepMind partners with Fenris Creations to tackle continual learning, deep memory, long-horizon planning and multi-agent dynamics in persistent game universes, aiming to discover new gameplay and apply insights to real-world problems.

Source: X (Grok x_search, @GoogleDeepMind) · within 48h · Original

13. Meta Previews WildArtifactBench for Multimodal Agents

AI at Meta previews WildArtifactBench, an evaluation framework using win rates and Elo from human/agent judges to assess multimodal agents on diverse real-world tasks, releasing 10 tasks to better measure practical utility.

Source: X (Grok x_search, @AIatMeta) · within 48h · Original

14. Elon Musk Highlights Grok Bot's Long-Running Agent Capabilities

Elon Musk showcases Grok Bot completing complex overnight coding, redesign and software stack tasks autonomously without babysitting, highlighting the shift to true agentic workflows where AI executes jobs independently.

Source: X (Grok x_search, @elonmusk) · within 48h · Original

15. Qwen3.8-27B Gets Major Quantization and Optimization Updates

Alibaba_Qwen highlights new Unsloth GGUF quants for Qwen3.8-27B with 10% higher accuracy and 1-bit options runnable on 8GB RAM, plus SGLang recipes for NVFP4 + DFlash2, boosting community efficiency.

Source: X (Grok x_search, @Alibaba_Qwen) · within 48h · Original

---

_Sources: official blogs of OpenAI, Google DeepMind and Qwen; official RSS of Simon Willison, Latent Space, Import AI and One Useful Thing; Anthropic site monitoring, top AI discussions on Hacker News and open-source updates from key AI orgs on GitHub. X sources are retrieved live via Grok x_search with per-link verification. This is a source digest — no commentary from Laojin._