Mercury 1
Inception · United States · 2025
Inception's first general-purpose diffusion model: 708 tokens per second measured by Artificial Analysis, about seven times the speed of the small models it matched on quality. Announced simply as Mercury in June 2025 and renamed Mercury 1 only once a successor existed.
Mercury 1 is the model with which Inception left the coding niche. Four months after Mercury Coder, on 7 June 2025, the company announced a general chat model built on the same diffusion architecture — a draft refined across all positions at once rather than written token by token. The question it had to answer was blunt: does the speed advantage survive when the task is no longer autocomplete inside a file, but open-ended conversation, general knowledge and maths? The answer, on the maker's own numbers, is a clear trade. Benchmarked by Artificial Analysis, an independent evaluator — which matters, because Inception's other speed claims are its own measurements — Mercury ran at 708 tokens per second against 96 for GPT-4.1 Nano and 67 for Claude 3.5 Haiku, the two small models it was positioned against. That is over seven times faster. On quality it landed roughly level with them and clearly below the larger models of the period: MMLU-Pro 69 percent, GPQA Diamond 51, HumanEval 85, MATH-500 83, LiveCodeBench 23, SciCode 18 and Humanity's Last Exam 3.4. Read together, the picture is a competent small model that answers almost instantly, not a frontier system. Two weaknesses are worth naming because they shaped what came next. Coding-heavy tests where the model has to reason through a problem rather than complete a pattern — LiveCodeBench and SciCode — were the model's worst results, below every competitor in the comparison. And the maker published no context window for this generation at all, which in a chat model is a real gap. Both were answered by Mercury 2 in February 2026: reasoning became the headline feature and the context window was finally stated at 128,000 tokens. The name is a small lesson in how product lines get written backwards. The announcement calls it simply Mercury, the general chat model; Mercury 1 is what Inception calls it today, retroactively, on the models page where it is listed among the older releases that remain supported for existing customers, with access and migration handled by a sales representative. In practice this means the model still runs and still bills, but nobody can buy it for the first time — the company now sells Mercury 2.5.
▸Videos
No videos yet.
▸Reviews
No reviews yet. Be the first!