Dispatch · 2 to 8 October 2026

GPT-6 in ChatGPT for everyone drew the most attention

OpenAI put GPT-6 in front of every ChatGPT user and Mistral previewed its largest model. OpenAI's mathematics results drew more Hacker News points than GPT-6, but no rise in Wikipedia reading.

Embed

The frame credits Hacker News via Algolia, Wikimedia Foundation, Hugging Face and YouTube and links back to its page. Preview it.

Cite
Download
Post on X

Two of five events cleared the bar

2 to 8 October 2026
01 GPT-6 in ChatGPT for everyone 96.7
02 Mistral Large 4 92.3
03 OpenAI publishes results on open problems in mathematics 62.6
04 Claude Haiku 5.5 58.4
05 DeepSeek Harness 57.6

Hacker News points · 40% Wikipedia spike · 30% Hugging Face upvotes · 20% YouTube views · 10% Dashed: the bar, 70. Dimmed: did not qualify.

This edition read five events, and two qualified. An event qualifies with a score of 70 or more from at least two signals.

Chart of the week

US data-center building rises 73% in a year, to $85.0 bn a year

Construction spending →

The two that cleared the bar

score out of 100
  1. 01

    GPT-6 in ChatGPT for everyone

    96.7

    OpenAI · 7 October 2026 · openai.com

    OpenAI started to roll out GPT-6 to every ChatGPT user, with Intelligent UI: answers that include visuals and interactive parts.

    Context ChatGPT is the largest consumer AI service, so a model change there reaches more people at once than any other release.

    Hacker News points
    748 points, 96.5th percentile · story
    Wikipedia spike
    9.3× its usual daily views, 99.8th percentile · GPT-6ChatGPT
    YouTube views
    512,263 views, 87.9th percentile · OpenAI
  2. 02

    Mistral Large 4

    92.3

    Mistral AI · 6 October 2026 · mistral.ai

    Mistral put Mistral Large 4 into public preview: a natively multimodal model with 1 trillion parameters, 52 billion of them active. Mistral says the open weights follow at the end of October.

    Context Mistral trained it on 3,800 NVIDIA Grace Blackwell GPUs in its own European data centers, and serves the preview from them.

    Hacker News points
    2,033 points, 99.9th percentile · stories 2,03352270
    Wikipedia spike
    1.4× its usual daily views, 82.2th percentile · Mistral AI
Also read, below the bar
Event Date Hacker News pointsWikipedia spikeHugging Face upvotesYouTube views Score
OpenAI publishes results on open problems in mathematics 6 Oct 1,326 points (99.2)0.8× its usual daily views (13.9)–– 62.6
Claude Haiku 5.5 7 Oct 1,039 points (98.3)0.7× its usual daily views (5.3)–– 58.4
DeepSeek Harness 2 Oct 417 points (89.7)0.8× its usual daily views (14.7)–– 57.6

New or refreshed on this site this week

11 datasets

Three papers

most upvoted on Hugging Face Daily Papers this week
  1. RealCompanion: Benchmarking Human Understanding from Reasoning over Longitudinal Real-World Conversations 276 upvotes · 5 Oct · arXiv 2610.01780
  2. OneStreamer: Unifying Perception, Memory, and Proactive Response in Streaming Video Interaction 235 upvotes · 2 Oct · arXiv 2610.01762
  3. On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics 199 upvotes · 2 Oct · arXiv 2609.35259

Said in public

“The big drop of this week was not a new AI model. It was instead the biggest day (so far!) in the history of mathematics”
Zvi Mowshowitz, thezvi.substack.com, 8 Oct
“It's certainly not a Fable-class model, but it's great to see Mistral put out a model that's back to being maybe about 6 months behind the frontier.”
Simon Willison, simonwillison.net, 6 Oct
“If your workloads fit in 100,000 tokens, Haiku is the same price as Luna and reports higher benchmark scores.”
Simon Willison, simonwillison.net, 7 Oct

Sources

scores.csv Every dataset → Status → Glossary →