Exploratory
A conversation and a look at the work. Early sight of public lab reports before wider publication.
Odin Labs R&D studies models, tools and vendors on bounded tasks and work from our own AI factory. We are looking for long-term partners to evaluate governed agents on their workflows, with agreed measures and evidence for each tested claim. The published experiments do not establish a daily delivery rate, autonomous company operation or customer outcomes at scale. Pick the track that fits you.
Corrected on
Earlier copy, corrected : “Odin Labs R&D is where models, tools and vendors meet real work. It is a running AI factory that ships changes every day, and a lab that measures what each part of that factory is worth. We are looking for long-term partners who share our view of where this is going: companies run by a governed factory of agents, with proof behind every step. Pick the track that fits you.” That wording stated goals and internal activity as established public outcomes. The published record does not establish those broader claims.
The same team and the same standard of evidence. What changes is what you bring and what you take away.
For model providers, brokers and compute platforms interested in an agreed evaluation on bounded factory workloads, with the method and publication terms set before a run.
data/provider-terms.json. Until then a provider is shown by letter.More for providers: what a provider gets and what we ask, and the brief, to print or forward.
For companies that want their own governed factory, following the Odin vision: agents that do real work, a record of every decision, and humans who approve what matters. A long-term partnership, not a project.
A conversation and a look at the work. Early sight of public lab reports before wider publication.
A completed evaluation or sprint and an agreement on the next. Throughput reporting for your own workflow.
Recurring work, plus your own cases or network brought in. Co-design of a benchmark or acceptance check.
We use the term only for things we can count. Where we cannot count it yet, we say so.
Merged-change throughput counts repository activity. It does not establish deployed functionality, accepted customer value or autonomy. A partner should see the measurement window and the separate acceptance evidence for its own workflow.
For teams with a hard question: a task they do not trust agents with yet, a risky migration, a security weakness, or a frontier direction. Selected tools and vendors can be considered within the agreed scope and terms.
Odin Labs reports an MIT application for Odin Next. This is author-reported; this page links no application or decision record and does not establish eligibility, approval, an award or the current decision status.
: earlier wording said the factory “ships changes every day”, with “proof behind every step”, offered measurement on “real work instead of a public benchmark”, and promised an identical rule set across providers. It also said the research “qualifies for innovation subsidies”. This page did not establish those general claims. The programme remains an offer to evaluate scoped workflows; repository activity, recorded evidence and an author-reported application are each narrower facts. How we qualify public claims.
: the earlier statement that “Every figure the lab has measured” is on the Evidence page implied a complete inventory. That page holds selected published claims and open questions.
It names no prices and no current or future partners, and it describes no signed partnership: none is signed yet. A partner is listed on the providers page only after an agreement is signed and on record.
Selected evidence and open questions links its displayed figures to published experiment records. It is not a complete inventory of the lab's measurements.
The evidence and the working method behind the programme.