Anthropic prepping new model that infers intent from minimal input, alongside Gemini 4 Carbon and OpenAI Terra
dotey · x · 2026-10-20
Per The Information, Anthropic is preparing a model with stronger reasoning and "intuition" than anything it currently ships, according to a person who has used it. Anthropic declined to comment.
- Better at unstated intent: given minimal input it infers the real goal — e.g., asked to fix a bug, it may trace out an entire class of related bugs and offer to fix them together.
- Fewer canned answers: asked about stock investing, it might design its own price-prediction algorithm rather than recite P/E-ratio boilerplate.
For users that means looser prompts; for developers, watch out for it proactively widening code changes. Name and launch timing are unconfirmed — TestingCatalog says this month, possibly in the Fable or Opus family. The same report notes Google is building Gemini 4 Carbon (an upgrade over last week's Gemini 4 Argon) and OpenAI a new Terra model with better capabilities at a lower price than GPT-6.1 Sol.
More from Models
- Exa 3.0 ties Perplexity at #1 on Artificial Analysis Search Index with score of 80 — yoimnotkesku · 2026-10-20
- Reddit user asks: do truly uncensored image models exist without a LoRA? — IngwiePhoenix · 2026-10-20
- 60 real PRs put Muse Spark Open to the test: 48% bug-fix rate, 72% clean PR approval — Arindam_1729 · 2026-10-20
- Mercury 3 diffusion LLM tested: matches Haiku 5.5 quality at nearly 2x the speed — daniel_mac8 · 2026-10-20
- MiniMax M3.1 Flash benchmark table leaks: tops automation evals at 62.8, size undisclosed — QuixiAI · 2026-10-20
- DiffusionGemma's blazing local speeds win fans for offline image-prompt agent — SysPsych · 2026-10-20