Inception's Mercury 3 diffusion model claims to beat GPT-6 Luna and Claude Haiku 5.5 at a fraction of the wait

AI startup Inception has released Mercury 3, a diffusion-based language model that refines a rough draft of the whole answer instead of writing word by word, which the company says makes it four to eight times faster than standard models.

• Inception calls Mercury 3 the most capable diffusion language model released so far, and says it generates 1,068 tokens per second on Nvidia GPUs, against about 240 for Claude Haiku 5.5 and 130 for GPT-6 Luna in the company's comparison.
• Inception's own benchmark table is more mixed: Mercury 3 wins eight of 12 tests at medium settings, leading IFBench with 76.84 points and SWE-Bench Atlas with 51.08, while Claude Haiku 5.5 takes the longer coding test DeepSWE and GPT-6 Luna edges ahead in two Artificial Analysis tests.
• No independent tests exist yet.
• The model handles up to 1 million tokens at once and uses an OpenAI-compatible API, and is available on Inception's own platform as well as OpenRouter, Baseten, and .