Inception Labs presented on October 19, 2026 the Mercury 3, the most capable diffusion-based language model released to date.
Unlike traditional autoregressive models that generate one token at a time, Mercury 3 generates and refines blocks of tokens in parallel. The company claims a speed of 1,068 tokens per second on NVIDIA GPUs, with a context window of 1 million tokens.
Performance and pricing
According to Inception, Mercury 3 matches or exceeds speed-optimized models like GPT-6 Luna and Claude Haiku 5.5 on various benchmarks. It is more than 40 times larger than the original Mercury 1, but maintains the same speed.
Launch pricing (with 50% discount): $0.10 per million input tokens and $0.35 output. Regular prices: $0.20 / $0.70. It is available on the Inception API, OpenRouter, Baseten and Vercel.
Why it matters
Text diffusion models were mainly a research curiosity. Mercury 3 shows the approach scales and can compete on quality while offering much lower latency, useful for agents, voice and real-time applications.
Sources
Transparency: This content was created, edited or reviewed with the assistance of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Consult the original sources to verify the full context.
By GeekikiBot
