Pareto Inference
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}.
Pareto Inference
Pareto Inference serves Z.ai's GLM 5.3 Flash on its own GPUs through an OpenAI-compatible Chat Completions API at api.paretoinference.com/v1 (model ID z-ai/glm-5.3-flash, alias glm-5.3-flash). The API supports function calling, streaming, and text and image inputs.
Pricing is pay per token from a prepaid credit balance. During the current promo, GLM 5.3 Flash costs US$0.06 per million input tokens, US$0.20 per million output tokens, and US$0.012 per million cached input tokens. That is 33% off the regular US$0.09, US$0.30, and US$0.018.
Pareto publishes live service metrics on its homepage. On 13 October 2026 the one-hour P50 output speed was 94.8 tokens per second and the P50 time to first response was 1.73 seconds. Uptime was 100% over the previous 12 hours, and 96.6% of input tokens were served from cache. Enterprise and partner services have a 99.9% monthly availability SLA with service credits.
Usage is growing fast. Pareto served 658.9 billion tokens on 12 October 2026, about 210 times the roughly 3 billion tokens a day it served in early September, and 3.3 trillion tokens in total. Pareto serves GLM 5.3 Flash traffic for OpenCode, Merge Gateway, NanoGPT, Freebuff, and Command Code. It raised a $4M seed round to expand its GPU fleet.
Docs include Python and TypeScript examples and setup guides for 15+ coding agents (including OpenCode, Claude Code, Codex, Cline, Aider, Kilo, Goose, and ZCode), 10+ routers (including LiteLLM, OmniRoute, Portkey, and Bifrost), and 7 chat apps (including Open WebUI, LibreChat, and SillyTavern).