Inception Labs has introduced Mercury 3, a diffusion-based language model the company says runs at 1,068 tokens per second on NVIDIA GPUs. Its claimed combination of speed and a one-million-token context window could make it relevant for AI services where waiting for an answer is part of the problem.
Watch Desk analysis
What happened
Inception says Mercury 3 is available through its API platform and through OpenRouter, Baseten and Vercel. The company describes it as more than 40 times larger than Mercury 1 while maintaining similar inference speeds, and says it matches or outperforms speed-optimised frontier models including GPT-6 Luna and Claude Haiku 5.5.
Those are the company’s claims in its Mercury 3 announcement. The supplied announcement excerpt does not give benchmark methodology or comparative results, so the headline speed and model comparisons need to be treated as Inception’s account, not an independent verdict.
Why it matters
A million-token context window and a claimed 1,068 tokens per second point towards latency-sensitive uses such as real-time search, browser agents, interactive education and customer support. If the performance holds up in practical workloads, developers could have another option for applications where a long context is useful but sluggish output is not.
Our read
This is a substantial model launch with a clear proposition: more context, quickly. The figures make Mercury 3 worth watching, but a speed claim is only as useful as the workload and comparison behind it. Developers should test their own tasks before treating the leaderboard language as a reason to switch.
What to watch
- Independent tests of speed and quality on comparable workloads.
- How Mercury 3 performs on tasks that use its full context window.
- Whether availability through the named platforms brings practical access for developers.
Discussion spark: For a real-time AI product, would you prioritise a million-token context window or independently demonstrated answer quality and latency on your own tasks?
Sources and evidence
- Diffusion at Scale: Introducing Mercury 3 – Inception (19 October 2026, 08:23 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.