Sports betting intelligence at a fraction of the cost
OpenAI-compatible models that know how bets settle, fetch live odds and scores mid-response, and read betslip screenshots into structured JSON. One grounding stack, two bases — the intelligence is in what we wrapped around them.
What you get in one API call
Every item below is produced by a specific mechanism — a rulebook we inject, a tool loop we run, a schema we enforce — not by hoping a bigger model knows betting.
Deep sports understanding
Football, racing, US sports, tennis, golf, cricket, combat, esports and more — with live odds, player props, scores, standings, results and rosters fetched mid-response by our gateway, not by your code.
The bookmaker rulebook, built in
Tattersalls Rule 4 bands, dead-heat division, place terms, pari-mutuel pools, exchange reduction factors — the conventions that decide what a bet pays, injected into every request.
Bet settlement that shows its working
Ask what a bet returns and it applies the right convention for the right jurisdiction, states the intermediate numbers, and gives you the figure.
Every bet structure
Singles, multis and parlays, same-game multis and same-game parlays — including SGMs nested inside multis — modelled as a tree of components and selections, not a flat guess.
Racing bets as a first-class citizen
Each-way with split stakes and dual odds, quinellas, trifectas, tricasts and exotics — plus proper race identity: venue, race number, or race time for UK and Irish cards.
Betslip reading and extraction
Send a screenshot from any bookmaker and get one canonical JSON back: every bet, leg, stake, odds, boost and status — 95.9% field-level accuracy on our 90-slip, 14-bookmaker benchmark — Lucky 15s, trifectas, quaddies and unplaced slips included.
Anomaly detection, not silent errors
The model checks the slip against its own arithmetic. When the numbers don’t reconcile — a dead heat, a Rule 4 deduction, commission — it flags the anomaly and names the candidates instead of forcing the maths.
A drop-in for what you already run
OpenAI-compatible chat completions with streaming, 1M-token context and vision. No tools array to build, no data-provider keys to hold, no prompt retention by default.
It knows how bets settle — and we can prove it
Extra time versus 90 minutes. A Rule 4 deduction keyed to the withdrawn horse’s price. A dead heat splitting the stake, not the odds. Place terms that change with field size. Every market settles by a rulebook, and general models have never been taught it — so we wrote the rulebook down, from the primary sources, and inject it into every Sports-1 request. When a convention changes, we edit a sentence, not retrain a model.
We grade it on SharpBench, the world-first AI benchmark for sports betting, odds and settlement: 93.6% overall, every settlement category at 87% or better — and on deduction maths it beats frontier models costing dozens of times more by over 30 points.
Live data without a data contract
Odds move and squads change after any model’s training cutoff. Sports-1 fixes that inside the request: when it needs a line, a live score, a table or a result, it calls our gateway’s tools — betting lines from named bookmakers including Pinnacle, live scores with the match minute, standings and form across every major sport — server-side, on our data contracts, with the result fed straight back into the same completion. Your integration is an ordinary chat request.
- 01
You send an ordinary chat completion
No tools array, no function-calling branch, no data-provider credentials. If your code speaks to OpenAI, it speaks to this.
- 02
The model fetches what it needs
Lines, player props, live scores, standings, results, rosters — the call never leaves our gateway, so there is no second round-trip through your process.
- 03
You get one answer, streamed
A normal completion with the live facts already inside. The grounding is invisible from the outside, which is the point.
Betslips in, structured JSON out
There are a thousand bookmakers and every app prints its slips differently, so Sports-1 doesn’t learn bookmakers — it learns the algebra every slip shares. A screenshot comes back as a tree: bets → components → selections. A same-game multi nested inside a parlay, an each-way with split stakes, a boosted price with the original struck through, three bets in one My-Bets screenshot — all just shapes of the same tree. Figures are reported as printed; when they don’t reconcile, the JSON says so and names the likely settlement cause instead of inventing numbers.
Measured at 98.5% field-level accuracy across a benchmark of real slips from eight bookmakers in two languages — singles, parlays, nested same-game parlays, exchange bets, each-way racing, futures and boosts, plus decoys it must refuse.
How to use it
Three ways in, from zero code to full integration. All of them are the same OpenAI-compatible endpoint underneath.
Hand it to your agent
Hermes, Openclaw, Opencode, Cursor, Claude Code — anything that accepts a custom OpenAI-compatible endpoint. Point it at api.infersia.com/v1 with model infersia/sports-1 and your agent can settle bets, quote live lines and read betslips today.
Build it into your product
Hand your coding agent the docs and it has everything it needs to wire deep sports understanding and betslip upload into your app — endpoint, schema, examples.
Or just call it
A plain chat completion is the whole integration — try it in the playground with your hardest settlement questions. Streaming, 1M context and vision work out of the box, and prompts are never retained by default.
Priced to run on every request
Sports-1 at $0.55 in / $0.85 out / $0.07 cached per million tokens, prepaid, with no subscription and no data contract. If you need a lookup or a settlement convention we don’t cover yet, tell us — the rulebook and the tool layer are the parts we extend on request.