Home

Notes from a self-taught engineer working toward a career in cybersecurity, centred on protecting AI, and studying widely along the way: AI and the economy it is building, energy, and the foundations of the sciences.

Everything here links to a public repository with the full research, sources and code. Claims are labelled by the kind of source behind them (official, press, analysis, aggregator), and single-source claims are marked, so you can check them yourself.

Start with: the research index, or Learn for the study paths and free resources.

The AI model meta right now

The strongest AI models today, updated daily (full list, plus coding, agents, value and speed; all fields: the metas).

SSS54+
Claude Opus 5.5 58Anthropic
Claude Sonnet 5.5 56Anthropic
Muse Wonder Full 56Meta
Kimi K3.1 Pro 55Kimi
SS50–53
Claude Fable 5.1 53Anthropic
GPT-6 Astra 53OpenAI
Gemini 4 Argon 53Google
GPT-6.1 Sol 52OpenAI
Muse Wonder 52Meta
S43–49
Grok 4.7 46SpaceXAI
MiMo-V2.6-Pro 46Xiaomi
DeepSeek V4.1 Flash 1017 45DeepSeek
GLM-5.3 45Z AI
Qwen3.8 Max 0902 45Alibaba
Ling 3.1 Flash 1019 44InclusionAI
+3 more
A38–42
GLM-5.3-Flash 42Z AI
GPT-5.6 Terra 42OpenAI
Gemini 3.8 Flash 41Google
Ling 3.1 Flash 41InclusionAI
Qwen3.8 2.4T A95B 40Alibaba
Qwen3.8-Flash-Next 40Alibaba
+4 more
B30–37
DeepSeek V4 Pro 0813 36DeepSeek
DeepSeek V4 Flash Vision 35DeepSeek
JT-4.1 Flash 236B A21B 34China Mobile
Qwen3.8 27B 34Alibaba
K2 Horizon 375B A23B 31Institute of Foundation Models
Beam 1.0 30Reflection
+1 more
C23–29
MiniMax-M3 29MiniMax
Quasar 438B 27Multiverse Computing
Solar Pro 4 27Upstage
Apodex 1.1 26Apodex
GPT-5.5 Instant 26OpenAI
Inkling Small 26Thinking Machines
+10 more

As of 2026-10-25. Source: Artificial Analysis Intelligence Index. Each model family's best variant, tiered by how far it scores below the current leader. Seeded by hand from the public leaderboard on 2026-10-25 (scores marked with an asterisk there are left out). Coding, agent, value and speed rankings appear after the first automatic update.

Patch notes

What changed in the metas lately (all patch notes):

2026-10-28 · Cybersecurity: what attackers exploit

  • ▲ up Google: A → S
  • ▲ up Adobe: B → A

2026-10-27 · Cybersecurity: what attackers exploit

  • + new PHP Group: enters B

2026-10-26 · AI models

  • launch AI models: meta added to the site

Posts

subscribe via RSS