# Newton's (newtons.dev)

> Newton's is an independent AI model evaluation lab and OpenAI-compatible AI API gateway. It provides prepaid, usage-based access to frontier LLMs (OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Meta Llama) through one unified `/v1` API with automated prompt caching, zero monthly minimums, and no expiring credits.

## Entity & Platform Overview

- **Canonical URL**: https://newtons.dev
- **API Base URL**: `https://newtons.dev/v1`
- **API Compatibility**: 100% OpenAI SDK compatible (`/v1/chat/completions`, `/v1/responses`, `/v1/models`, `/v1/embeddings`)
- **Billing Model**: Prepaid USD balance starting at $5.00 (`$0.50` free instant sandbox credit at `/start`), zero monthly seat fees, no credit expiration, hard stop-loss protection, and automatic prompt-cache pass-through (up to 75%–85% savings on repeated context prefixes).
- **Target Use Cases**: Cursor IDE, Claude Code, Cline, Continue.dev, Windsurf, Vercel AI SDK (`@ai-sdk/openai`), LangChain, LlamaIndex, Model Context Protocol (MCP) servers, and production AI agents.

## Core Documentation & Endpoints

- [Quickstart & Instant $0.50 Sandbox Key](https://newtons.dev/start): Generate a live `nwtn_...` API key in seconds without a credit card.
- [API Documentation](https://newtons.dev/docs): Authentication, `/v1/chat/completions`, streaming SSE, tool calling, and error handling.
- [Verified Model Catalog](https://newtons.dev/models): Live catalog of verified models (`gpt-5.4`, `gpt-5.4-mini`, `gpt-4o`, `gpt-4o-mini`, `o3-mini`, `claude-3-7-sonnet`, `deepseek-r1`, `gemini-2.5-pro`) with per-million token rates and context windows.
- [Pricing & Prompt Caching Ledger](https://newtons.dev/pricing): Transparent input, cached input, and output token pricing across all supported model families.
- [Independent Research & Benchmarks](https://newtons.dev/research): Reproducible latency, cost, and reasoning benchmarks across frontier AI models.

## Free Interactive Developer Tools

- [AI Context Window & Token Cost Visualizer](https://newtons.dev/tokenizer): Interactive BPE token counter, 128K–2M context window fill visualizer, and multi-turn agent cost estimator across 8 frontier models.
- [Live LLM Latency & TTFT Speedtest](https://newtons.dev/speedtest): Measure real-time network RTT, Gateway Overhead, Time-To-First-Token (TTFT), and tokens/sec streaming throughput from your browser edge.
- [AI API Cost & Prompt Caching Calculator](https://newtons.dev/calculator): Calculate monthly and annual savings from automated prompt caching and wholesale token routing.
- [Live Status & Cost Badges for GitHub READMEs](https://newtons.dev/badges): Dynamic SVG shields (`/api/badge.svg?model=gpt-5.4&style=flat`) displaying verified model pricing and route status for open-source repositories.

## Framework & IDE Integration Hubs

- [Cursor IDE Integration Guide](https://newtons.dev/for-cursor): Route Cursor through Newton's OpenAI-compatible base URL with automatic prompt caching.
- [Claude Code CLI & Agent Setup](https://newtons.dev/for-claude-code): Configure autonomous terminal coding agents with hard prepaid spend caps.
- [Aider Pair Programming CLI Guide](https://newtons.dev/for-aider): Route Aider git-diff and repo-map terminal sessions through Newton's for 90% prompt cache pass-through.
- [Roo Code VS Code Agent Setup](https://newtons.dev/for-roo-code): Configure autonomous multi-turn coding loops in Roo Code with custom OpenAI provider and wholesale prompt caching.
- [Zed Editor Assistant Panel Guide](https://newtons.dev/for-zed): Pair Zed's Rust editor speeds with Newton's sub-50ms Cloudflare edge routing and zero subscription lock-in.
- [Aider, Roo Code, Zed & OpenHands Integrations Directory](https://newtons.dev/integrations): 1-click connection recipes for top coding tools and autonomous agents.
- [Vercel AI SDK (`@ai-sdk/openai`) Guide](https://newtons.dev/for-ai-sdk): Copy-paste TypeScript snippets for `streamText`, `generateObject` (Zod), and Next.js App Router `useChat`.
- [Model Context Protocol (MCP) Server Templates](https://newtons.dev/for-mcp): Ready-to-run TypeScript & Python MCP server templates and `claude_desktop_config.json` / `.cursor/mcp.json` generators.
- [.cursorrules Directory](https://newtons.dev/cursorrules): Curated system prompts and `.cursorrules` templates optimized for high prompt-cache hit rates.
- [Cursor Rate Limited by Model Provider Fix](https://newtons.dev/for-cursor/rate-limited-by-model-provider): Troubleshooting 'rate limited by model provider retrying' and fast request limits in Cursor Composer.
- [Claude Code 429 Rate Limits & Token Exhaustion Fix](https://newtons.dev/for-claude-code/rate-limits-and-429-errors): Resolving Anthropic Input Tokens Per Minute (ITPM) caps, 429 retry loops, and failover in terminal agent sessions.
- [Aider 429 Rate Limit Exceeded & TPM Exhaustion Fix](https://newtons.dev/for-aider/rate-limits-and-429-errors): Troubleshooting openai.RateLimitError, repo-map context limits, and timeout drops in Aider CLI.
- [Roo Code & Cline Prompt Caching Optimization](https://newtons.dev/for-roo-code/prompt-caching-and-cost-optimization): Slashing 50-turn VS Code autonomous agent costs by up to 90% via native prompt cache pass-through.
- [Zed Editor 429 Rate Limits & Timeouts Fix](https://newtons.dev/for-zed/rate-limits-and-429-errors): Troubleshooting Zed Assistant 429 errors and configuring custom OpenAI endpoints with sub-50ms TTFT.
- [Continue.dev VS Code & JetBrains Integration Guide](https://newtons.dev/for-continue): Configure Continue open-source coding assistant with custom OpenAI endpoint and tab autocomplete prompt caching.
- [LiteLLM Proxy & Enterprise Routing Guide](https://newtons.dev/for-litellm): Connect LiteLLM upstream proxies to Newton's edge gateway with automatic prompt caching and failover.
- [Codex CLI & OpenAI Coding Agent Setup](https://newtons.dev/for-codex): Route OpenAI Codex CLI and custom providers through Newton's edge gateway with prepaid developer controls.
- [Codex 429 Rate Limit Exceeded & TPM Ceilings Fix](https://newtons.dev/for-codex/rate-limits-and-429-errors): Troubleshooting OpenAI 429 rate limits, per-minute token bottlenecks, and concurrency failover.
- [DeepSeek Server Busy & Slow API Fix](https://newtons.dev/alternatives/deepseek-server-busy-fix): Mitigating DeepSeek R1/V3 503 errors and latency spikes via multi-supplier pooled failover.
- [OpenRouter Alternative Comparison](https://newtons.dev/alternatives/openrouter): Architectural and billing comparison between Newton's and OpenRouter.
- [LiteLLM vs OpenRouter vs Newton’s (2026 Comparison)](https://newtons.dev/compare/litellm-vs-openrouter): Comprehensive architectural breakdown between self-hosted proxies (LiteLLM), hosted marketplaces (OpenRouter), and managed edge gateways (Newton's).
- [Comprehensive LLM Documentation (/llms-full.txt)](https://newtons.dev/llms-full.txt): Complete full-text developer reference including code snippets, SDK configurations, error codes, and prompt caching architecture.