GoKawiilInception Labs launches Mercury 3 diffusion-based language model
Inception Labs released Mercury 3, a text-diffusion large language model it says outperforms GPT-6 Luna and Claude Haiku 5.5 on various benchmarks while generating over 1,000 tokens per second. The model supports a 1-million-token context window, up from 260K in Mercury 2.5, and is priced at $0.20 per million input tokens and $0.70 per million output tokens, discounted 50% at launch. Early-access partners are reportedly using it for chat support, tutoring, data engineering, and web-browsing agents.
GoKawiil's interpretation of the reporting above, not reported fact.
The company frames Mercury 3 as evidence that diffusion-based language models can scale in capability much like autoregressive models have, which could open a faster, cheaper alternative architecture for latency-sensitive AI applications. If these speed and quality claims hold up under independent testing, it may pressure established providers of 'speed-optimized' models to respond on both performance and pricing.
- Mercury 3 claims to beat GPT-6 Luna and Claude Haiku 5.5 on benchmarks while running 4-8x faster.
- The model offers a 1-million-token context window and runs at over 1,000 tokens per second on NVIDIA GPUs.
- Inception Labs says Mercury 3 is 40x larger than Mercury 1, suggesting diffusion LLMs can scale similarly to autoregressive models.
Source: inceptionlabs.ai, 2026-10-19
Published there as: “Mercury 3: Diffusion at Scale”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.