magnitude

Open-source inference server for local model routing

devtoolsadvancedFree tier

Overview

magnitude is an open-source inference server that routes requests to the most suitable local model based on available hardware. Developers should care because it simplifies local model serving and can sit underneath existing agent tooling without major workflow changes.

Best For

+ml-engineers
+platform-engineers
+backend-developers

News & Updates

Developers Get 99% Cheaper Doc Extraction as Pacific Claims 3x Speed Over Frontier

Pacific released a document extraction tool that it claims delivers near-frontier structured output quality at up to 99% lower cost and 3x faster speed. The company says the tool 'owns the Pareto curve for almost two orders of magnitude,' meaning it dominates the cost-quality tradeoff across a wide range. Pacific has been using the extraction engine internally under the hood, suggesting it's production-tested rather than a lab demo. The announcement targets developers who currently pay premium prices for structured data extraction from documents like invoices, contracts, and forms.

breakthroughs10/20/2026

Nscale and Anthropic Eye Second $45B Compute Deal After August Pact

Nscale and Anthropic are reportedly discussing a second compute deal worth $45 billion, matching the size of their August agreement. The August deal was already one of the largest AI infrastructure contracts ever signed, and a second deal of the same magnitude would double Anthropic's committed compute spend with Nscale. That level of commitment suggests Anthropic is securing massive GPU capacity to train and serve future Claude models, likely beyond its current needs. Nscale, a UK-based AI cloud provider, has now become a critical infrastructure partner for one of the world's top AI labs. The talks signal that compute scarcity is driving multi-billion-dollar, multi-year contracts between AI model developers and specialized cloud providers.

deals10/19/2026

Claude Mythos 5 Designs Quantum Circuits That Outperform Published Records Across Eight Primitives

Anthropic reported on October 16, 2026 that its Claude Mythos 5 model, operating in a closed loop with an automated evaluator and largely on its own, designed quantum circuits that outperform the best previously published designs across eight computational primitives. In some cases, the improvements were by an order of magnitude or more. This marks a significant advance in AI-driven scientific discovery, showing that large language models can autonomously generate novel quantum algorithms. The closed-loop system allowed Claude to iteratively refine its designs without human intervention, potentially accelerating quantum computing research.

breakthroughs10/16/2026