Mercury 3
Inception’s flagship diffusion LLM, with a 1M-token context.
Mercury 3 is Inception’s flagship diffusion language model. Like earlier Mercury models it refines many tokens in parallel instead of writing them one by one; Inception reports over 1,000 tokens per second on NVIDIA GPUs (vendor-reported). It raises the context window to 1M tokens and offers four reasoning-effort levels, from instant to high.
- Provider
- Inception
- Type
- Language model
- Released
- Oct 19, 2026
- Context window
- 1M tokens
- Max output
- 131K tokens
- Price
- $0.10 input / $0.35 output per 1M tokens
- Input
- text
- Output
- text
- Open weights
- No
Best for
- Real-time / low latency
- High volume / low cost
- Coding
- Agents
- Long context
Strengths
- Parallel (diffusion) generation
- 1M-token context
- Low price
More from Inception
- Mercury 2.5 — Inception. Diffusion language model that refines many tokens in parallel.
- Mercury Decide — Inception. Inception’s structured decision endpoint: choices, scores and yes/no with probabilities.