{"id":1609,"date":"2026-06-22T04:38:58","date_gmt":"2026-06-22T04:38:58","guid":{"rendered":"https:\/\/xrpfaucet.site\/?p=1609"},"modified":"2026-06-22T04:38:58","modified_gmt":"2026-06-22T04:38:58","slug":"inception-labs-mercury-2-ai-beats-googles-diffusiongemma-at-its-own-game","status":"publish","type":"post","link":"https:\/\/xrpfaucet.site\/?p=1609","title":{"rendered":"Inception Labs’ Mercury 2 AI Beats Google’s DiffusionGemma at Its Own Game"},"content":{"rendered":"
\n

\ud83d\udcf0 Exclusive Crypto News & Analysis:<\/strong> Stay ahead with the latest developments in the cryptocurrency and blockchain space.<\/p>\n

\ud83d\udcc8 Market Update:<\/strong> Real-time price movements, technical analysis, and trading signals.<\/p>\n

<\/p>\n

\n
\n

In brief<\/h4>\n
    \n
  • Inception Labs’ Mercury 2 generates roughly 1,000 tokens per second and scored 90 on the AIME 2026<\/li>\n
  • Google’s recent DiffusionGemma hits similar speeds but performs worse on benchmarks.<\/li>\n
  • DiffusionGemma is free and open-weight on Hugging Face. Mercury 2 is a paid, closed-weight API model.<\/li>\n<\/ul>\n<\/div>\n

    Inception Labs introduced Mercury 2 on Thursday, calling it the world’s fastest reasoning language model. Per the company’s announcement, it generates about 1,000 tokens per second\u2014the chunks of text an AI model reads and writes\u2014against roughly 89 tokens per second for Anthropic\u2019s Claude Haiku 4.5 Reasoning and 71 for OpenAI\u2019s GPT-5 Mini.<\/p>\n

    That puts it in the same speed bracket Google would later claim for BroadcastGemma<\/a>.<\/p>\n

    \n
    \n

    Welcome to the diffusion era. <\/p>\n

    We bet on parallel generation years ago, when it was a contrarian idea. It’s great to see the industry arrive. <\/p>\n

    Mercury 2 continues to lead the Pareto frontier for quality, speed, and cost among publicly available diffusion LLMs. pic.twitter.com\/qSHuiR7vmH<\/a><\/p>\n

    \u2014 Inception (@_inception_ai) June 18, 2026<\/a><\/p>\n<\/blockquote>\n<\/div>\n

    Both models get there by dropping the typewriter approach to writing. A standard chatbot writes one word, checks what it just wrote, then writes the next, looping until the answer is finished. Diffusion models instead fill a block of text with random placeholder tokens and erase the noise across a handful of parallel passes\u2014the same trick that turns static into a photo in image generators like Stable Diffusion\u2014until the whole block locks into a finished response at once.<\/p>\n

    Where the two diverge is what survives that process. On AIME 2026\u2014built from real American Invitational Mathematics Examination problems and scored as the percentage solved correctly\u2014Mercury 2 hit 90%. Google tested DiffusionGemma on the same set, where it scored 69.1%, while standard, non-diffusion Gemma 4 scored 88.3% on the same test.<\/p>\n

    On GPQA, a PhD-level science benchmark scored the same way, the two models nearly tie: Mercury 2 at 77% against DiffusionGemma’s 73.2%. But Google’s own developer guide recommends standard Gemma 4 for applications that demand maximum quality, conceding DiffusionGemma trails it across the board.<\/p>\n