Mercury 3 lands on AI Gateway: diffusion-based parallel refinement generates text above 1,000 tokens per second
Synopsis
Inception's Mercury 3 is now available through AI Gateway, generating text by diffusion through parallel refinement rather than word-by-word sequential generation. Inception reports strong results across a range of benchmarks, generation speeds above 1,000 tokens per second, and a 1M-token context window. The model targets latency-sensitive workloads such as browser agents, real-time search, interactive education, and customer support, with improved reasoning.
Interpretation
Mercury 3 can now be called through AI Gateway under the model name inception/mercury-3, with launch pricing at 50% off. The model was not previously among the models available on AI Gateway, so callers needed another route to reach it. AI Gateway offers a unified API for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations; it reflects provider pricing with no markup and charges no platform fee on inference.
Mercury 3 generates text with diffusion, producing output through parallel refinement instead of advancing one token at a time. Conventional text generation emits tokens in sequence, whereas this approach recasts generation as repeated parallel refinement. Inception reports strong results across a range of benchmarks and generation speeds above 1,000 tokens per second.
A 1M-token context window lets long documents and extended conversation histories fit within a single request. Latency-sensitive applications were previously constrained by context length when holding long material and long conversations together. The model is positioned for workloads such as browser agents, real-time search, interactive education, and customer support, and brings improved reasoning to them.
Perspective
The report covers the AI Gateway listing and the capabilities Inception describes: diffusion-based parallel refinement, generation speeds above 1,000 tokens per second, a 1M-token context window, and positioning for latency-sensitive workloads such as browser agents, real-time search, interactive education, and customer support. Access is by setting the AI_GATEWAY_API_KEY environment variable and using the model name inception/mercury-3, or by running the Vercel CLI ai-gateway setup to select the model in coding agents such as Claude Code, Codex, and Cursor.
The "strong results across a range of benchmarks" that Inception reports come without benchmark names or numbers in the material, and the speed figure comes without stated measurement conditions. The improved reasoning and the fit for latency-sensitive scenarios are the releaser's descriptions; the material gives no measured performance in those scenarios.