TogetherLink routes Claude Code and Codex to open models in beta

Auto-routing sends routine prompts to cheap open models, hard Claude work to Opus. Plus: Cloudflare's agent web search and a diary that got a user arrested.

Nowline OCT 6 11:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • One installer, your agents on open models

    TogetherLink is a beta CLI that connects Claude Code, Codex, OpenCode 2, Pi Code, Claude Desktop, and ChatGPT Desktop to Together-hosted open models. Install with one curl command, then launch via `tclaude`, `tcodex`, or `topencode`.

  • Auto-routing: open by default, Opus when it's hard

    A virtual `auto` model classifies each prompt — routine work goes to fast open models, while in Claude Code or Desktop (with an Anthropic key added) the hard tasks route to Claude Opus and bill to your own account. Codex, OpenCode, Pi, and ChatGPT Desktop always stay on Together models.

  • The model menu and what it costs

    Kimi K3 runs $2.70 in / $13.50 out per 1M tokens with a 1M-token context; GLM 5.3 is $1.40 / $4.40; Qwen 3.8 $2.00 / $6.00; DeepSeek V4.1 Flash just $0.30 / $1.20. That's a fraction of frontier pricing for the routine 80% of your coding session.

  • The catch: beta, and not actually free

    Despite early write-ups calling it free, only the installer and router are — usage is billed per-token at Together's standard serverless rates, and you bring your own Together key. It's in beta, so commands, routing behavior, and the model list may still change.

  • Cloudflare ships a Web Search API for agents

    Cloudflare's new beta Web Search API lets agents ground answers in live results via Exa, Linkup, or Ceramic, with zero data retention. It's billed through AI Gateway at each provider's list price with no markup — or plug in your own provider key.

  • A Claude 'diary' got a user arrested

    A Florida woman's Sept 26 entries threatening a sheriff's office were flagged by Claude's safety systems, escalated by a human reviewer, and reported to police; she now faces a second-degree felony. A reminder for anyone building on these tools: flagged content isn't private — Anthropic's policy permits disclosure to prevent death or serious injury.