Inception Mercury 3, a new diffusion-based language model, is now available, and the company says it generates text up to eight times faster than traditional AI models. Co-founder Stefano Ermon announced the release on October 19, 2026. Coverage of the launch has set the model against ChatGPT, but the eight-times figure is stated against traditional AI models, and the company has not published the benchmarks behind its wider quality claims.
How the speed claim is built
Most large language models produce text one word at a time. Mercury 3 takes a different route: it refines several blocks of text at once, which Inception likens to an editor revising a whole draft in one pass. The company says this lets the model produce more than 1,000 tokens per second on widely available NVIDIA graphics processors. A token is a small unit of text that an AI model reads and writes.
Inception also says Mercury 3 beats GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks. It has not yet named those benchmarks or released scores, so independent testing has not confirmed the claim. The model's context window, meaning how much text it can handle in one session, has grown to 1 million tokens from 260,000 in Mercury 2.5. Inception further says Mercury 3 is more than 40 times larger than the original Mercury model while keeping comparable processing speeds.
Availability and what it is built for
Inception Mercury 3 can be used through Inception's own API and through Baseten, OpenRouter and Vercel. The model supports the OpenAI API format, which the company says lets developers connect it without major code changes. Inception aims it at tasks where delay hurts, including interactive tutoring, live translation, writing assistants and automated software agents.
Early-access customer RevisionDojo reported that Mercury 3 cut student feedback response times to roughly one second. That figure comes from one customer and has not been independently measured. The Mercury 3 AI model follows Mercury 1, launched in June 2025, and Mercury 2, which added reasoning. The company has raised $50 million from investors including Microsoft, NVIDIA and Snowflake, and Ermon confirmed that Inception has already started training its next model.
Diffusion-based models still face doubts about long-term competitiveness. Google's Gemini Diffusion showed similar promise, but it has had no public updates since its first presentation. Whether such models can keep pace with fast-improving autoregressive rivals, the one-word-at-a-time designs that dominate the field, remains an open question.
Timeline: what has happened so far
- June 2025: Inception launches Mercury 1.
- Since June 2025: Mercury 2 adds reasoning, and Mercury 2.5 adds adjustable reasoning with a 260,000 token context window.
- October 19, 2026: Stefano Ermon announces the release of Inception Mercury 3.
Aasan alfaaz mein
Inception Mercury 3 ek naya AI model hai jo company ke mutabiq traditional AI models se 8 guna tez text banata hai. Yeh 19 October 2026 ko launch hua, aur iski speed aur benchmark ke daway abhi independent tor par verify nahi hue. Aam log ke liye iska matlab hai ke future mein tutoring, translation aur writing apps zyada tez jawab de sakti hain. Lekin abhi yeh sirf company ka claim hai, is liye ChatGPT se tez hone ka faisla abhi jaldi mein na karein.
