Closed-to-open model transfer · Founder-led engagement
Distill expensive
traces. Own the
model.
Converra takes one workload you run on a frontier model and tests whether a small open model can do the same job. You get the evidence either way.
One workload. One answer.
- 01
Picks one workload
The expensive, repetitive one. One task, and the quality bar your team already holds it to.
- 02
Builds the test from your traces
Saved runs become the question set. Your team confirms which answers count as correct. A protected set is held back before any training.
- 03
Compares before it trains
A small open-weight model answers the protected set next to the model you run today, untuned. If it falls short, it trains on the reviewed runs and answers again.
- 04
Returns a verdict
Pass, fail, or not enough evidence. A pass earns a trial on live traffic, and you approve the switch.
Is it your workload?
A fit
- One narrow, repetitive task.
- Busy enough to pay for its own serving.
- The evidence an answer needs is already in the prompt, the tools or the retrieved documents.
- A correct answer can be checked.
Not a fit
- Open-ended judgment that leans on the closed model's own knowledge or taste.
- Low volume.
- A contract that changes every week.
- Answers nobody on your team can mark right or wrong.
What you bring
- The agent's exact contract: instruction, tools, output schema, current model.
- Saved runs that carry their evidence, not final answers alone.
- One named reviewer with hours each week to decide what counts as correct.
- A written bar: what must pass, and by how much.
- The rights to use your data for a model only you will hold.
What you get.
- What you own
- The model, any adapter and the reviewed dataset. They are yours under our Terms and are never shared with another customer.
- What trains it
- Your own saved runs, reviewed by your team. Insights point out what went wrong; they never stand in for a correct answer.
- How you know
- Both models answer the same protected questions and are scored the same way. You see every answer, not only the total.
- Where it runs
- Your provider account or ours, decided with you before anything is trained. Providers are barred by contract from training on your data.
- How it is priced
- A paid diagnostic on one workload comes first, at a fixed fee agreed before anything is trained. A scoped sprint follows only if the test says go.
- Who does the work
- Oren Cohen, Converra's founder, runs every engagement: scoping, data review, training and the verdict.
What we don’t promise.
- That the small model passes. Some workloads need the frontier model. When that is the answer, the test says so.
- A savings number before the test. Serving cost depends on your volume. At low volume, a dedicated small model can cost more than the API you use now.
- A production result from a lab score. A passing test earns a trial on live traffic. Real users decide the rest.
From production · Verified on real traffic
22 → 0
Invented pricing, contact emails and product features, across 97 sampled conversations after the fix.
That result came from a prompt fix, not a model transfer. It is here because the same production verdict judges a model switch.
The method, in public
48 parts on moving a production agent from a closed model to an open one.
Data selection, leakage, judging, serving cost and rollback. Written from the runs that failed as well as the ones that held.
Read the series →Questions buyers ask.
- What is closed-to-open model transfer?
- Closed-to-open model transfer moves one workload from a closed frontier model, such as the API you call today, to a small open-weight model that you own and can run where you choose. Converra runs it as a test on one workload first: the small model answers a protected set of your own reviewed runs next to the model you run today, and that result decides whether the workload moves.
- Do I need to fine-tune the small model?
- Not necessarily: the small model answers the protected set untuned first, and it trains on your reviewed runs only if it falls short. A trained model has to earn its place on the same protected questions.
- Which workloads are a fit for model transfer?
- A narrow, repetitive task with enough volume to pay for its own serving, where the evidence an answer needs is already in the prompt, the tools or the retrieved documents, and where a correct answer can be checked. Open-ended judgment that leans on the closed model's own knowledge or taste is not a fit.
- What do I need to provide before the test starts?
- Five things: the agent's exact contract (instruction, tools, output schema and current model), saved runs that carry their evidence, one named reviewer with hours each week to decide what counts as correct, a written bar for what must pass, and the rights to use your data for a model only you will hold.
- How is the engagement priced?
- A paid diagnostic on one workload comes first, at a fixed fee agreed before anything is trained, and a scoped sprint follows only if the test says go. Serving cost afterwards depends on your volume; at low volume a dedicated small model can cost more than the API you use now.
- Who owns the model and the training data?
- You do: the model, any adapter and the reviewed dataset are yours under Converra's Terms and are never shared with another customer. Training runs in your provider account or ours, decided with you first, and providers are barred by contract from training on your data.
- What does a passing test mean?
- A pass means the small model met the bar your team set on the protected questions. It is not a saving and it is not a production result. A pass earns a trial on live traffic, where real users decide, and you approve the switch.
Start with one workload.
Thirty minutes to pick the workload and the bar. You see the test plan and the fixed fee before anything is trained.