OpenRouter reveals stealth 'Union Alpha' as Pareto, a paid 262K model

The free model devs were quietly probing is now paid at $2.50/M — plus a free 2M-token code router, agent skills from Notion, and drop-in Mem0 memory.

Nowline SEP 18 7:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Union Alpha, out of stealth

    The anonymous free model builders were hammering on OpenRouter is now live as Pareto from The Unbiased Co: a multimodal (text + image in) composite model with a 262K context and 131K max output. OpenRouter confirmed the two are the same model — but The Unbiased Co ships no model card, parameter counts, or benchmarks, so treat the 'frontier-level' billing as unverified for now.

  • Free while stealth, $2.50/M now

    Coming out of stealth flipped the meter on. Pareto bills $2.50/M input and $7.50/M output, with cached input at $0.25/M. If you wired anything to Union Alpha's free endpoint over the past two days, those calls now cost money — swap the model ID or set a budget before your next run.

  • A free code router you tune by score

    The same shop shipped Pareto Code Router — free, with a 2M-token context — which dispatches to a shortlist of top coding models ranked by Artificial Analysis percentiles. Set min_coding_score from 0 to 1 to trade cost for strength, or add the Nitro variant to rank by throughput instead of quality.

  • Install agent skills from a Notion doc

    Vercel's skills CLI can now pull agent skills straight from a Notion database — no Git repo required — so a team can keep its skill library where it already writes docs. Landing alongside it: Mem0 is now a native Vercel Marketplace integration, giving your agents and apps memory that persists across sessions.

  • Elsewhere: Google's CC becomes a family agent

    Google Labs relaunched CC as a shared AI agent for up to six household members, each with their own private context and controls. Consumer-facing, but a marker of where multi-seat agent accounts — and the sharing model behind them — are heading.