Pay-as-you-go inference API for open-weight models from a GPU-kernel lab, with zero data retention by default; DeepSeek V4.1 Flash from $0.14 input and $0.58 output per million tokens, $1 free credit.
Isoquant
New· 4An inference cloud for open models where research agents keep tuning the serving stack to each customer's workload; public prices from $0.035 per million input tokens, dedicated endpoints on request.
- Artificial Intelligence - AI
- Machine Learning
- API
- Cloud
TrustList summary
Not yet independently verified. Dated from the Y Combinator launch post (16 November 2026, UTC). The speed, token and customer figures are the maker's own statements from that post and have not been checked independently. Prices are from the maker's site, read 18 November 2026; the maker's site says its SOC 2 audit is in progress. We will update this when it can be confirmed, and remove this note.
Running an open model with stock serving software usually leaves a lot of the hardware idle. Isoquant, a Y Combinator company from the Fall 2026 batch, runs an inference cloud in which its own research agents keep testing and changing the serving stack for each customer's workload.
How the tuning works
A customer brings its tasks and the measures of success it cares about. Isoquant's agents then run experiments on the whole serving path, which the launch post lists as GPU kernels, memory, decoding, caching and scheduling, and keep the winning settings. The site describes the same loop for training small specialist models, and shows case studies in which a tuned small model matched or beat a larger one on a banking, coding and returns-policy task. Those results are the maker's own.
Public endpoints and prices
The site lists two public models, with prices per million tokens for input, cached input and output. DeepSeek V4.1 Flash 1017 Fast costs $0.035, $0.005 and $0.37. GLM-5.3-Flash costs $0.07, $0.025 and $0.15. Dedicated endpoints that are tuned for one customer's workload are quoted through the sales team.
The maker's claims
The launch post says Isoquant serves open models about four times faster than other providers, above 1,200 tokens a second with latency under 500 milliseconds, and that after one month it was serving more than 1.5 trillion tokens for over 1,000 customers, growing 40% a week. These are the maker's statements. The site says its SOC 2 audit is still in progress.
Key facts
Maker: Agent FM Inc. (Y Combinator, Fall 2026) Launched: 16 November 2026 (Y Combinator launch post) Product: self-optimising inference cloud for open models Public prices per million tokens: DeepSeek V4.1 Flash 1017 Fast $0.035 input, $0.005 cached, $0.37 output; GLM-5.3-Flash $0.07, $0.025, $0.15 Dedicated endpoints: by quote Security: SOC 2 in progress (maker's site) Website: isoquant.ai
Categories & features
- Artificial Intelligence - AI
- Machine Learning
- API
- Cloud
Trust Score
An earned signal from verification, reviews, awards, transparency and engagement - the vendor can't buy it.
Updated 11/18/2026
Request a demo or quote from Isoquant
Reviews
Questions & answers
No questions yet. Be the first to ask about Isoquant.
Isoquant alternatives
Similar listings buyers compare against this one.
Training data made to order for AI labs: speech in nine languages, video, image and text collection, annotation on its own Foundry platform, and reinforcement-learning environments and benchmarks.
Reinforcement learning environments in which live data streams in on a clock while an AI agent works, covering personal agents, industry operations and science and compute, for training models to reason over changing signals.
Small on-device AI models for speech, text and vision, delivered as SDKs for mobile, web and Node
Optimisation platform that benchmarks prompts, models and agents on each team's own data, with tracing, for AI product teams
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- Write a reviewHelp other buyers decide
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ArticlesGuides, comparisons and research from the editorial desk and the community.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.