Skip to content
Research
Model release

PredictLM API: self-serve access is live

The hosted PredictLM API is now self-serve — register, get a key in seconds, and call zero-training tabular prediction over HTTPS. Now with calibrated uncertainty, feature attributions, and a plain-language explanation in one call.

Zero One Research2 min read·#predictlm#api#release#explainability

Six weeks ago we opened early access to the hosted PredictLM API and onboarded users one conversation at a time. Today that gate comes down: the PredictLM API is self-serve. Create an account, issue an API key, and you're making predictions in under a minute — no waiting, no sales call, a free tier to start.

Get your API key →

What changed

The early-access API answered one question — what's the prediction? The live API answers three, in a single call:

  • What — the prediction itself. Float targets route to regression, integer targets to classification with calibrated per-class probabilities.
  • How sure — a calibrated uncertainty distribution (mean, standard deviation, quantiles, an 80% interval), not just a point estimate.
  • Why — ranked feature attributions, plus an optional plain-language rationale that narrates the prediction grounded strictly on those numbers.

You choose how much detail you want, per call:

POST /v1/predict     → just the prediction          (fast, 1 unit)
POST /v1/explain     → prediction + uncertainty + feature attributions
                       (+ a written rationale, if you ask for it)        (5 units)

If you don't need the narration, request "want": ["attributions"] on /v1/explain and you get the calibrated uncertainty and attributions with zero LLM tokens.

Quickstart

# 1. register and grab a key
curl -s https://predictlm-api.zerooneresearch.ai/auth/register \
  -H 'content-type: application/json' \
  -d '{"email":"you@company.com","password":"a-strong-password"}'
# -> {"access_token":"...", "plan":"free"}
 
# 2. predict (X-API-Key from the dashboard, or the Bearer token above)
curl -s https://predictlm-api.zerooneresearch.ai/v1/predict \
  -H 'X-API-Key: ml_sk_...' -H 'content-type: application/json' \
  -d '{"X_train":[[5.1,3.5,1.4],[6.2,2.9,4.3]],"y_train":[0,1],"X_query":[[5.9,3.0,5.1]]}'

One small training table in, predictions out — in-context learning in a single forward pass. No fit, no hyperparameter sweep, no model selection.

What it runs

Under the hood it's predictlm-base-26m, the Apache-2.0 checkpoint already on Hugging Face. The weights stay free — that hasn't changed. The API is for the cases where you don't want to own inference: spreadsheets and internal tools, agents using prediction as a tool call, and teams that want the current-best model without tracking our releases. When a stronger model clears our public-benchmark and latency gates, it swaps in behind the same contract.

Free while in research preview

The hosted API is free for now — no pricing tiers, no card, no billing. Every account is self-serve and instant, and includes a monthly allowance of calls (1,000 units a month; a prediction uses 1 unit, an explanation 5 — enough to integrate and ship a small tool). Need a bigger allowance? Tell us what you're building and we'll raise it. You can watch live usage any time in the dashboard.

Create an account → · prefer to talk first? Get in touch.

More from research