# Mercury 3 Launches

> Inception's Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 at 1,068 tokens per second.
- Title: Mercury 3 Launches · Inceptionlabs | Shortify
- Summary: Inception's Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 at 1,068 tokens per second. Inception's Mercury 3 is the most capable diffusion-based LLM…
- Keywords: inceptionlabs, mercury 3, diffusion llm, gpt-6 luna, claudia haiku 5.5, nvidia, tokens per second, ai model, technology, startups, Mercury, Launches
- Source: Inceptionlabs — https://inceptionlabs.ai/blog/introducing-mercury-3
- Published: 2026-10-19T00:00:00+00:00
- Read time: 1 min
- Topics: inceptionlabs, mercury 3, diffusion llm, gpt-6 luna, claudia haiku 5.5, nvidia, tokens per second, ai model, technology, startups
## Mercury 3 Launch
Inception introduces Mercury 3, the most capable diffusion-based LLM released to date.
## Performance Metrics
Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 while running 4-8x faster at 1,000+ tokens per second.
## Scaling Success
Mercury 3 is more than 40x larger than Mercury 1 and demonstrates text diffusion scales.
## Early Access Partners
Early-access partners use Mercury 3 for voice and chat support, real-time tutoring, data engineering, and web navigation.
## Mercury 3 at a Glance
Quality: On par or better than frontier speed-optimized models like GPT-6 Luna and Claude Haiku 5.5.
## Speed Metrics
1,068 tokens per second on NVIDIA GPUs with 1M token context, up from 260K for Mercury 2.5.
## Pricing Details
Price: $0.20 per million input and $0.70 per million output tokens at launch 50% off.
## Browser Use Agents
Mercury 3 completes 50% of Mind2Web navigation tasks faster than GPT-6.1-Sol with same token volume.
## Interactive Education
Mercury 3 provides instant feedback to students with quality as good as or better than previous models.
## Real-time Search Agents
Mercury 3 delivers median generation time of 0.9 seconds and end-to-end response time of 2.2 seconds.
## Key takeaway

Inception's Mercury 3 is the most capable diffusion-based LLM, outperforming GPT-6 Luna and Claude Haiku 5.5 while running 4-8x faster at 1,000+ tokens per second.