Coding agents mostly spend compute on hyperparameter tuning, rarely attempting the algorithmic research that make human records successful.
In one instance, Codex spent 121 H100 hours adjusting two values in the training code: cooldown fraction and window size schedule parameters.
@Yuchenj_UW available capacity is growing on Mithril (@mithrilcompute) since people can pause reserved instances and lend them to spot pools. spot pools always have capacity at dynamic prices (which are low during off-peak times)
A thoughtfully designed benchmark is catalytic for research progress.
At @mithrilcompute, we've partnered closely with @nikogrupen, @gabepereyra , @ItsJulioPereyra, and the @harvey team on LAB, with a focus on understanding agent performance and optimizing sub-agent delegation across models for long-horizon legal work.
This is exactly the kind of benchmark the field needs: realistic, client matter-centric, and grounded in expert evaluation.
More to share about our joint work soon.
Mithril is opening new large "Flexible Reservation" H200 and B300 clusters in EMEA.
Reach out if interested.
Customers leveraging these regions can pause instances and make them available to others via spot, recouping costs when not used.
This mechanism is demonstrating 30%+ cost offsets so far for paused capacity.
This is why Mithril built "Flexible Reservations". Users can pause instances and feed compute back to the grid, so to speak, letting nodes into the spot pool when they aren't using them and earn cash-back when they aren't used. This disincentivizes faking utilization.
Lastly, our sponsor @mithrilcompute made this workshop possible!
Some of us use their platform for high-performance computing and we love the experience!
Mithril is offering compute credits (1x $10,000 org, 2x$1,000 indiv)
Apply here by May 1st: https://t.co/XaDxf1yJJr
#ICLR2026
We're working with Google to bring next-gen TPUs to Mithril's Omnicloud as well. Mostly current and former Google/DeepMind researchers appreciate TPUs today.
Excited to make it more seamless for the broader ecosystem to experience them.
It's an honor to work for and with the cleverest, most innovative teams developing new fundamental methods in ML and applying AI to new frontiers in physical intelligence, AI for science, and more.
Proud of what the team is building. If compute economics and ML systems work excite you, Mithril is hiring!
We are hiring at @mithrilcompute!
Access to high-performance compute (GPUs, TPUs, and the like) is fundamentally broken: 1. prohibitively expensive and 2. heavily under-utilized.
The providers (mostly publicly traded companies) want financial guarantees that the bet will pay off: contracts with them must be large and long. Only a small set of players (i.e., well-funded "startups") can afford those terms.
The demand from these players is incredibly high—Blackwell GPUs are essentially sold out everywhere—but if you look at the utilization numbers, something is way off. Why?
At Mithril, we believe this is due to a fundamental fact: price control just doesn't scale with AI usage patterns. Small and medium players want access to these advanced chips in unpredictable patterns:
"I need to immediately run this experiment for 48 hours and then again 15 days from now" said a genomics researcher somewhere.
"My inference service explodes in demand every end of the month; I need 500 more chips just for 3 days every month" said the CEO of an Accounting AI platform somewhere else.
This is an incredibly rich field to be part of! At Mithril, you’ll be working to enable millions of research institutions and companies worldwide to leverage one of the pinnacles of human ingenuity—without the usual hassle.
From building our advanced Virtual Machine and high-performance network orchestration, to creating durable, resilient APIs and establishing state-of-the-art consumption principles... there’s just too many GOOD open problems to solve!
Take a look at our openings at https://t.co/Jtl1kwwQXf.
All positions are in-person at our offices in San Francisco and Palo Alto, CA.
Customers are using Mithril "Flexible Reservations" and capacity relist extensively!
A few hundred additional Blackwells and Hoppers, which would otherwise be allocated but underutilized, have been added back to the spot pool over the last week or so.
As a result, you can get 1x-8x B200 instances for $0.01/gpu/hour on Mithril. We expect availability to continue to expand, and for this to play a role in resolving the industry-wide capacity crunch.
Today, we come out of stealth. 👋
@urunml is the inference cloud for the interactive era.
We wrote down why we're building it and what we believe.
Manifesto→ https://t.co/sbYPyYopfX
Join the waitlist → https://t.co/yTmInev13V
TPUs are coming to Mithril! Mithril GPU spot already starts at $0.01/gpu/hr (not a typo) for A100s-B200s.
TPU self-serve reservations now available, and spot pricing is next.
Algorithmic spot pricing = cheap when others aren't using it. Nights and weekends are basically free. Schedule accordingly.
Great to be partnered with Nebius!
Any @nebiusai customer can enable Mithril-Nebius to use Mithril's tools and turn any standard reservation into a flexible reservation that they can pause to earn:
Computer use models shouldn't learn from screenshots.
We built a new foundation model that learns from video like humans do. FDM-1 can construct a gear in Blender, find software bugs, and even drive a real car through San Francisco using arrow keys.
Computer use models shouldn't learn from screenshots.
We built a new foundation model that learns from video like humans do. FDM-1 can construct a gear in Blender, find software bugs, and even drive a real car through San Francisco using arrow keys.
We've opened new pools for on-demand, SPOT, and self-serve reserve (arbitrary durations, from hours to weeks) NVIDIA B200 GPUs on Mithril.
In general, these chips are hard to get access to, so we hope this helps!
Spot floor at $0.01 for long-running and flexible jobs.
Blackwells are really nice to work with. Having all the extra HBM is super convenient.
Researchers at the Broad Institute use Mithril’s GPU omnicloud to better understand gene expression.
Read how the Broad Institute is leveraging Mithril to accelerate biological discovery in our latest case study.
https://t.co/A9nVSywxbw