Inception releases Mercury 3, a diffusion LLM claiming 1,000+ tokens per second

Inception launched Mercury 3, a diffusion LLM with a 1M-token context, OpenAI-compatible API and vendor-reported speed and benchmark gains.

Inception released Mercury 3 on October 19, 2026. It is a diffusion LLM that the company says outperforms GPT-6 Luna and Claude Haiku 5.5 on a range of benchmarks while running 4–8x faster, at 1,068 tokens per second on NVIDIA GPUs. If latency is your bottleneck, as in voice agents, chat support or agent loops that call a model many times, this is the fast-tier model to test this month. The quality and speed numbers are Inception's own, though. TNW notes that no outside group has published a test yet.

The specs: a 1M-token context (up from 260K on Mercury 2.5), and list pricing of $0.20 per million input tokens and $0.70 per million output. At launch both are halved, to $0.10 and $0.35. TechRepublic reports that the announcement gives no end date for the discount. The API is OpenAI-compatible and available on Inception's platform, OpenRouter, Baseten and Vercel. New keys get 100 million free tokens, per Inception's models page.

Measured traffic is more mixed. An eesel review cites OpenRouter percentiles from October 20: 221 tok/s at the median, but 79 tok/s or slower for one request in ten. It also flags some API differences:

So pin medium before you compare quality against your current model.

If you run Mercury 2 or 2.5 today, Inception says access for existing customers ends November 15, 2026. Enterprise customers stay supported under their agreements.

Sources