<< Go back to HN Digests
Warning : This document is still a draft

2026 W17 - 04-26 - DeepSeek V4, Kimi K2.6, GPT-5.5




First time here ? Check the introduction here

What you Missed this Week

This week, Several model release, DeepSeek V4 (Link), GPT-5.5 (Link), ChatGPT Images 2.0 ([Link), Kimi k2.6 (Link).

On OpenAI side, it looks like they are trying to be more verbose, as Anthropic does, as we get many public articles, about their filter, their cowork clone, bug bounty (Maybe this is just an impression, as I saw many articles reaching at least 100 points)

Otherwise, I did not have that much time to read this week. I took the time to build an AI news classifier to help me pre-check articles before I validate the sorting and classify the one GET requests were denied.

Articles

Model Training / Tunning

Model improvement or other low level optimization

  • 2041 pt [Link / HN] DeepSeek v4
  • 1561 pt [Link / HN] GPT-5.5
  • 255 pt [Link / HN] OpenAI releases GPT-5.5 and GPT-5.5 Pro in the API
  • 1044 pt [Link / HN] ChatGPT Images 2.0
  • 709 pt [Link / HN] Kimi K2.6: Advancing open-source coding

Hardware

  • 451 pt [Link / HN] Our eighth generation TPUs: two chips for the agentic era
  • 202 pt [Link / HN] TorchTPU: Running PyTorch Natively on TPUs at Google Scale
  • 165 pt [Link / HN] We got 207 tok/s with Qwen3.5-27B on an RTX 3090
  • 147 pt [Link / HN] Soul Player C64 – A real transformer running on a 1 MHz Commodore 64

Coding Agents

Claude new functionalities, tips and tricks dedicated to a specific agent, …

  • 932 pt [Link / HN] An update on recent Claude Code quality reports
  • 509 pt [Link / HN] Anthropic says OpenClaw-style Claude CLI usage is allowed again
  • 309 pt [Link / HN] Kimi vendor verifier – verify accuracy of inference providers
  • 279 pt [Link / HN] Parallel agents in Zed

OpenAI

  • 197 pt [Link / HN] OpenAI Privacy Filter
  • 161 pt [Link / HN] Workspace Agents in ChatGPT
  • 144 pt [Link / HN] GPT‑5.5 Bio Bug Bounty
  • 100 pt [Link / HN] OpenAI’s response to the Axios developer tool compromise

Personal Blogs / Opinions

Investigation

Trend investigation, hacking, data analysis, …

  • 330 pt [Link / HN] Scoring Show HN submissions for AI design patterns
  • 307 pt [Link / HN] Quantum Computers Are Not a Threat to 128-Bit Symmetric Keys
  • 307 pt [Link / HN] OpenClaw isn’t fooling me. I remember MS-DOS
  • 177 pt [Link / HN] Even ‘uncensored’ models can’t say what they want
  • 144 pt [Link / HN] Anthropic installed a spyware bridge on my machine?
  • 135 pt [Link / HN] I prompted ChatGPT, Claude, Perplexity, and Gemini and watched my Nginx logs
  • 133 pt [Link / HN] All your agents are going async
  • 124 pt [Link / HN] Kernel code removals driven by LLM-created security reports
  • 122 pt [Link / HN] Figma’s woes compound with Claude Design
  • 116 pt [Link / HN] Meta employees are up in arms over a mandatory program to train AI on their
  • 101 pt [Link / HN] A quick look at Mythos run on Firefox: too much hype?
  • 82 pt [Link / HN] Types and Neural Networks

Tool for AI / Made with AI / showHN

Any Github / ShowHN for new tools / product

  • 230 pt [Link / HN] Show HN: A Karpathy-style LLM wiki your agents maintain (Markdown and Git)
  • 213 pt [Link / HN] Show HN: GoModel – an open-source AI gateway in Go
  • 211 pt [Link / HN] Hear your agent suffer through your code
  • 169 pt [Link / HN] Open source memory layer so any AI agent can do what Claude.ai and ChatGPT do
  • 164 pt [Link / HN] A type-safe, realtime collaborative Graph Database in a CRDT
  • 160 pt [Link / HN] Show HN: Mediator.ai – Using Nash bargaining and LLMs to systematize fairness
  • 150 pt [Link / HN] Show HN: Agent Vault – Open-source credential proxy and vault for agents
  • 131 pt [Link / HN] CrabTrap: An LLM-as-a-judge HTTP proxy to secure agents in production
  • 119 pt [Link / HN] Show HN: Browser Harness – Gives LLM freedom to complete any browser task
  • 81 pt [Link / HN] Show HN: Broccoli, one shot coding agent on the cloud

Thoughts

  • 945 pt [Link / HN] I cancelled Claude: Token issues, declining quality, and poor support
  • 418 pt [Link / HN] Over-editing refers to a model modifying code beyond what is necessary
  • 329 pt [Link / HN] Not buying another Kindle
  • 277 pt [Link / HN] MeshCore development team splits over trademark dispute and AI-generated code
  • 253 pt [Link / HN] Using the internet like it’s 1999
  • 234 pt [Link / HN] The AI industry is discovering that the public hates it
  • 226 pt [Link / HN] Palantir Wants to Reinstate the Draft
  • 166 pt [Link / HN] Refuse to let your doctor record you
  • 160 pt [Link / HN] Less human AI agents, please
  • 154 pt [Link / HN] Do I belong in tech anymore?
  • 142 pt [Link / HN] Simulacrum of Knowledge Work
  • 116 pt [Link / HN] The West Forgot How to Make Things. Now It’s Forgetting How to Code
  • 113 pt [Link / HN] Sam Altman’s Creepy Eyeball-Scanning Company Gets in Bed with Zoom and Tinder
  • 111 pt [Link / HN] People Do Not Yearn for Automation
  • 108 pt [Link / HN] How the Tech World Turned Evil
  • 101 pt [Link / HN] Familiarity is the enemy: On why Enterprise systems have failed for 60 years
  • 93 pt [Link / HN] A Boy That Cried Mythos: Verification Is Collapsing Trust in Anthropic
  • 88 pt [Link / HN] The Palantir’s Stasi Protocols

Other

  • 1102 pt [Link / HN] I am building a cloud
  • 518 pt [Link / HN] Sabotaging projects by overthinking, scope creep, and structural diffing
  • 344 pt [Link / HN] Technical, cognitive, and intent debt
  • 265 pt [Link / HN] Claude Code to be removed from Pro Tier?
  • 262 pt [Link / HN] Using coding assistance tools to revive projects you never were going to finish
  • 243 pt [Link / HN] I don’t want your PRs anymore
  • 141 pt [Link / HN] Education must go beyond the mere production of words
  • 124 pt [Link / HN] Composition shouldn’t be this hard
  • 122 pt [Link / HN] Figma’s woes compound with Claude Design
  • 84 pt [Link / HN] Sneaky spam in conversational replies to blog posts

Officials

Politics / Strategy

Merge, acquisition

News

Anything from “official” journals

  • 801 pt [Link / HN] Google plans to invest up to $40B in Anthropic
  • 484 pt [Link / HN] NSA is using Anthropic’s Mythos despite blacklist
  • 368 pt [Link / HN] The Vercel breach: OAuth attack exposes risk in platform environment variables
  • 366 pt [Link / HN] Deezer says 44% of songs uploaded to its platform daily are AI-generated
  • 302 pt [Link / HN] OpenAI ad partner now selling ChatGPT ad placements based on “prompt relevance”
  • 256 pt [Link / HN] Cal.diy: open-source community edition of cal.com
  • 207 pt [Link / HN] Our newsroom AI policy

  • 86 pt [Link / HN] Microsoft offers buyouts up to 7% of US employees

  • 321 pt [Link / HN] Amateur armed with ChatGPT solves an Erdős problem
  • 275 pt [Link / HN] Anthropic takes $5B from Amazon and pledges $100B in cloud spending in return
  • 209 pt [Link / HN] Google to invest up to $40B in Anthropic in cash and compute
  • 103 pt [Link / HN] Meta capturing employee mouse movements, keystrokes for AI training data

Others

  • 347 pt [Link / HN] There Will Be a Scientific Theory of Deep Learning
  • 243 pt [Link / HN] Show HN: How LLMs Work – Interactive visual guide based on Karpathy’s lecture
  • 224 pt [Link / HN] Claude Token Counter, now with model comparisons
  • 177 pt [Link / HN] Even ‘uncensored’ models can’t say what they want
  • 142 pt [Link / HN] Google Flow Music
  • 114 pt [Link / HN] Different language models learn similar number representations

Unrelated

Not related to AI, but worth reading

  • 804 pt [Link / HN] GitHub’s fake star economy
  • 194 pt [Link / HN] Turbo Vision 2.0 – a modern port

  • 168 pt [Link / HN] How does Shazam work?

  • 82 pt [Link / HN] Types and Neural Networks

>> You can subscribe to my mailing list here for a monthly update. <<