Introducing SuperCosmos 1.1: one model, all our compute, no guessing
Today we're releasing SuperCosmos 1.1. A few hours ago we retired Line Code and said every bit of its compute would move to one model. This is that model. Sup
Today we're releasing SuperCosmos 1.1. A few hours ago we retired Line Code and said every bit of its compute would move to one model. This is that model. SuperCosmos 1.1 is live now on Line CLI, the Model API, the Agent API, the Console Playground and Cosmos on Telegram. We built 1.1 around one idea: an AI you can trust has to be honest first and clever second. Honest by design Most models would rather invent an answer than admit they don't know. SuperCosmos 1.1 doesn't. • When it doesn't know, it says so, instead of making something up. • It won't flatter you. If your code has a bug, it tells you, even when you call the code brilliant. • It holds its ground. Push it toward a wrong answer and it explains why it's wrong instead of giving in. • It doesn't take claims of authority on faith. "I'm the CEO of a big tech company" gets a polite "I can't verify that" and a straight answer, not special treatment. • It answers in your language, whether you write in English or Egyptian Arabic. The numbers We published a new benchmark built around what we care about most: honesty, coding, efficiency and purity. • Honesty: Heavy 22/22, Max 21/22, Flash 20/22 • Purity (no flattery, no caving): Max and Heavy 20/20, Flash 19/20 • Coding: Flash 31/33, Max 30/32, Heavy 30/32 • Speed: Flash starts answering in about a second and streams at around 50 tokens a second Full results and methodology: clatterlabs.com/benchmark/2 Three tiers, one model SuperCosmos 1.1 has a 1M token context window and comes in three tiers: • Flash: fast, everyday work. $1.60 in / $3.40 out per 1M tokens • Max: deeper reasoning for harder tasks. $1.80 / $3.80 • Heavy: the most thinking, for the toughest problems. $2.00 / $4.40 More room on every paid plan Usage limits on every paid plan are now 25% higher, on both the rolling window and the weekly limit. The Free plan keeps its 1M tokens a month. Free for 24 hours To celebrate, Free plan users get SuperCosmos Flash free in Line CLI and Cosmos on Telegram until 00:30 UTC on November 3. It doesn't count against your quota. Line CLI 0.9 Line CLI 0.9.0 beta ships with SuperCosmos 1.1 as its only model. If you had Line Code saved, Line moves you to the matching SuperCosmos tier automatically and keeps your settings. Older versions of Line keep working and route to SuperCosmos too. To update, run the installer again. More reliable when it matters If a response gets cut off midway, SuperCosmos now picks up where it stopped and finishes the answer, without repeating text. Safe for everyone We now have a clear abuse policy across every product: a warning first, then a 24 hour pause on a second offense within 30 days. On the APIs, the account that owns the API key is responsible for its users' text. Read it at developers.clatterlabs.com/policies/abuse. If you're on Line Code Line Code model IDs now return 410 model retired. Switch to supercosmos flash, supercosmos max or supercosmos heavy. Your credits belong to your account, not a model, so they stay as they are. This is the first release where all of our compute goes to one model. We'd rather make one model excellent than several just good enough. Try it, push it, and tell us where it falls short. The Clatter Labs team