Inception · v2 · since 24. Februar 2026 (Ankündigung); OpenRouter nennt 4. März 2026 · 2× · last seen Oct 21, 2026
26
Momentum
Mercury 2 is a reasoning language model from Inception based on diffusion architecture instead of sequential token generation. It is offered via an OpenAI-compatible API and supports tunable reasoning levels, a 128K context window, tool use and schema-aligned JSON output. Inception positions it for latency-sensitive applications such as voice assistants. Note: per Inception's model page, Mercury 2 is no longer available to new users; Mercury 3 is the successor.
Momentum trend
24.07.22.10.
Features
| Real-Time Streaming | Yes: block-by-block streaming via OpenAI-compatible chat completions (stream: true) |
| Latency | 1,009 tokens/s on NVIDIA Blackwell; TTFT <300 ms on standard NVIDIA GPUs (vendor claim) |
| Platform | Inception API (OpenAI-compatible), OpenRouter, Azure AI Foundry, Vercel AI Gateway, Puter.js |
| Price | $0.25 / 1M input tokens, $0.75 / 1M output tokens; cache read $0.025 / 1M |
| Release Date | February 24, 2026 (announcement); OpenRouter: March 4, 2026 |