Skip to content

AI model

Mercury

Mercury is a diffusion-based large language model from Inception Labs that generates text through parallel denoising passes instead of token-by-token decoding.

Current stories

investOne report1 publisher

Mercury alone drew more chatbot recommendations than all banks combined in AIVO's small-business test

ChatGPT, Gemini and Perplexity recommended fintechs 1,345 times and banks 452 in 2,160 chats with researchers posing as small-business owners. Because the final prompt asked how to open an account, banks that still require a branch visit face a gap that tagging their websites may not close.

Reality

Evidence55
Adoption
Not scored
Hype gap+30
Incentives
Not scored
Confidence50
investOne report1 publisher

Community banks added $482 billion of deposits while stablecoins grew

The American Bankers Association warns that as much as $6.6 trillion of transactional deposits is exposed to stablecoin migration. That pool is about 2.6 times what every US bank under $10 billion holds in total, and small-bank deposits grew anyway.

Publishers:fortune.com

Reality

Evidence55
Adoption25
Hype gap+12
Incentives65
Confidence48
scienceConfirmed3 publishers

Roughness maps push Mercury's total contraction up by 10 to 30 percent

Mercury's lumpiest terrain carries the fewest shrinkage wrinkles, which the authors read as impact debris burying the record. Correct for that and total contraction rises to as much as 23 km, with the thermal models still to be redone.

Perspective Coverage

3 publishers
Builder
Builder 42%
Operator
Operator 33%
Investor
Investor 25%

Reality

Evidence64
Adoption18
Hype gap+22
Incentives40
Confidence66
scienceOne report1 publisher

Half-millimetre glass beads push Mercury's lava source deeper into its mantle

A team calibrated infrared spectra against beads of known silica content, checked the method on the Moon where returned rocks can grade it, and now reads Mercury's surface as up to 25 percent poorer in silica than assumed.

Reality

Evidence55
Adoption
Not scored
Hype gap+18
Incentives45
Confidence48
buildOne report1 publisher

Masked diffusion sampling spends whole-sequence passes to escape the token-by-token loop

Mercury reports over 1,000 tokens per second per user, and Nemotron Diffusion claims 2 to 8 times autoregressive throughput. Both are vendor figures. The sampler that produces them charges for its speed in arithmetic.

Publishers:dev.to

Reality

Evidence38
Adoption42
Hype gap+30
Incentives62
Confidence40