Mercury 3
Inception Labs released Mercury 3, a diffusion LLM it says matches GPT-6 Luna and Claude Haiku 5.5 on benchmarks while running 4 to 8 times faster at 1,068 tokens per second.
Mercury 3 has more than 40 times the parameters of Mercury 1, and Inception says benchmark intelligence rose 3x from Mercury 1 mini. Its context window is 1M tokens, up from 260K for Mercury 2.5. Inception says this shows scaling laws hold for text diffusion models.
- Date
- Monday 19 October 2026
- Lab
- Inception Labs
- Kind
- model
- Access
- closed API
- Price
- $0.20 per million input and $0.70 per million output tokens; launch price 50% off at $0.10 input and $0.35 output; cached tokens 90% off input price
Figures
| Measure | Value | Measured by |
|---|---|---|
| Speed | 1,068 tokens per second on NVIDIA GPUs | company |
| Context window | 1M tokens up from 260K for Mercury 2.5 | company |
| Speed vs comparison models | 4-8x faster versus GPT-6 Luna and Claude Haiku 5.5 | company |
| Mind2Web live navigation tasks completed | 50% 50 live tasks in early-access partner TinyFish AI's production web-agent harness, reported as significantly more than GPT-6.1-Sol; quoted on Inception's blog | company |
| Scale vs Mercury 1 | more than 40x parameters | company |
Available on the Inception API Platform, OpenRouter, Baseten and Vercel. Benchmarks are company reported. Inception also mentions Mercury Voice and Mercury Decide.
Sources
This record was checked and corrected against its sources on 19 October 2026. How we check