Meta's Muse Spark 1.2 lands on OpenRouter — from $0.10/M input
The model's been out since Aug 5; what's new is global access, a rock-bottom Contributor tier that trades cost for your data, and Meta's muse CLI.
Copy markdown
One key away, 1M context
Muse Spark 1.2 is now live on OpenRouter with expanded global access, so you can route to it — with automatic fallbacks — using the key you already have. Standard pricing is $1.25/M input and $4.25/M output ($0.15 cached), with a full 1M-token window and text, image, video, audio, and PDF input.
The $0.10 Contributor tier — and its catch
A Contributor tier cuts input to $0.10/M ($0.002 cached) and output to $0.20/M, but Meta gets to train on your prompts and completions, and you're capped at 60 requests per minute versus 3,000 on standard. Ideal for experiments and learning; keep production and anything sensitive on the standard tier.
Tuned for agent loops
The model does structured output, parallel function calling, and configurable reasoning effort, and Meta aims it at multi-file refactors, whole-repo generation, and long debugging runs. Vendor-run numbers: 82.9% on Terminal-Bench 2.1 and 59.3% on DeepSWE — self-reported, with no independent reproduction yet.
Drive it from the terminal with muse
Muse Code, Meta's terminal agent (the `muse` binary, macOS and Linux, one-line install), defaults to Muse Spark 1.2 and adds background agents, worktree-isolated sub-agent fan-out, and crash-resume from an append-only event log. It's bring-your-own-key, so you can aim it at other models too.
Elsewhere: Anthropic's Python SDK v1.0 can break your build
The new anthropic 1.0 needs Python 3.10+, swaps httpx for httpx2, and drops the legacy Text Completions API plus temperature/top_p/top_k on Messages; AnthropicBedrock now errors if no AWS region is set instead of defaulting to us-east-1. Pin your version before CI trips on it.
Elsewhere: Vercel starts billing deployment storage
New Pro and Enterprise teams now pay $0.10/GB per month for deployment storage, with retention controls added to Deployment History. Deploy often? Prune old builds or watch the line item grow.