🎓Task: Deep research on recent analysis of the impact of AI on the economy
Mercury 3 + Mercury Decide beats GPT-6 Luna on speed *and* quality
- Mercury 3 plans, interprets, and synthesizes
- Mercury Decide filters for relevance
(What were the conclusions of the reports? Read
Mercury 3 is at the Pareto frontier of speed vs intelligence, reaching over 1000+ tokens/sec.
In: $0.10 / 1M. Out: $0.35 / 1M
Try it today: https://chat.inceptionlabs.ai/ (🧵 2/2)
Coming soon to San Francisco International Airport? In this demo from @harwiltz, Mercury 3, our flagship diffusion-based LLM, outperforms Claude Sonnet 5.5 and GPT-6 Luna on simulated air traffic control (🧵 1/2)
Victory in 2 min 51 secs! In this @StarCraft-esque demo from @hjing28, Mercury 3 quickly conquers GPT-6 Luna.
Our new model, Mercury 3, outperforms GPT-6 Luna and Claude Haiku 5.5 while running 4-8x faster at 1,000+ tokens per second. (🧵1/2)
Mark Peng from @Tiny_Fish in an interview with @_inception_ai's Emily Liu: “We’ve been running the latest Mercury 3 diffusion model through our most mature evaluation harness at TinyFish for the web agent and the results have been quite impressive.
In terms of task completion
Mercury 3 excels in tasks where a combination of strong intelligence and speed is necessary - in this example it is an order of magnitude faster and cheaper than competing models.
Try it today: https://docs.inceptionlabs.ai/get-started (🧵2/2)
Yesterday, we launched Mercury 3, our fastest model to date with speeds over 1000+ tokens/second on @nvidia hardware.
In this demo from @stefan_webb, we compare Mercury 3 + @opencode against Claude Code with Sonnet 5.5 and other model/harness combinations on a coding task
A clarification: “Intelligence Index” refers here to a weighted average of the following benchmarks: GDPval, SciCode, Tau3-Banking, AA-Omniscience-Public, IFBench, SO Quality, AA-LCR v1.1, GPQA Diamond, AA-Omniscience-Public (Non-Hallucination), Terminal-Bench v2.1, BrowseComp+,
Mercury 3 is live: the most capable diffusion LLM released to date.
It outperforms GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks while running at over 1,000 tokens per second.
More than 40x the size of Mercury 1, 3x more intelligent across standard benchmarks,
Mercury 2 has been named one of @TIME's Best Inventions of 2026!
Our first reasoning diffusion LLM, and the first of its kind in production, was recognized in the AI category.
Thank you to the judges for recognizing diffusion LLMs as one of the technologies shaping the future
Mercury Decide with vision performs strongly on vision benchmarks, leading CV-Bench outright and beating the current 2nd place, Imajev-4B, on the public set of Image JevBench.
How do you get started? (🧵3/4)
Mercury Decide with vision is useful in many applications of visual understanding and decision making outside of real-time gameplay. Some examples include:
- Verify from a screenshot that a browser use agent succeeded after taking an action
- Tag millions of images with
👁️ Mercury Decide now accepts image input, which opens it up to a much wider range of applications:
https://openrouter.ai/inception/mercury-decide:free
In the demo from @stefan_webb, Mercury Decide plays Ring King on NES, using Mercury Decide’s new vision capabilities to understand the state of the game
