Sports-2

The sibling of Sports-1: Infersia’s proprietary sports grounding stack — the live-data tools, the settlement rulebook, the betslip extraction schema — on a 27-billion-parameter dense vision base instead of Sports-1’s sparse mixture-of-experts. Exposed side by side so you can compare them on your own traffic. Model id infersia/sports-2.

Calling the model

Everything in the Sports-1 reference applies here unchanged — automatic grounding with no tools array to build, live odds and scores fetched server-side mid-response, settlement questions run tool-free for exact arithmetic, harness callers keep their own tools and system prompt, and betslip screenshots come back in the same canonical JSON schema. The only change is the model id.

the whole integrationbash

curl https://api.infersia.com/v1/chat/completions \
  -H "Authorization: Bearer $INFERSIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "infersia/sports-2",
    "stream": true,
    "messages": [
      {"role": "user", "content": "What are the current lines for Chiefs vs Bills?"}
    ]
  }'

What’s different from Sports-1

Parameter	Type	Description
`Base model`	`27B dense`	A 27-billion-parameter dense vision model: all 27B parameters compute on every token. Sports-1 is a 320B sparse mixture-of-experts activating 18B per token.
`Pricing`	`higher output`	Dense compute is what the higher output price pays for — see the model page for current per-token rates. Input and cached-input prices stay in the same range as Sports-1.
`Max output`	`32k tokens`	Per response. Context window is 1M tokens, the same as Sports-1.
`Long context`	`text-only`	Requests near the top of the context window are served text-only; keep image requests (betslip extraction) to ordinary conversation sizes.
`Positioning`	`comparison arm`	Here to be compared, not to replace Sports-1. Same stack, different base — run both on your own workload and keep whichever settles your bets better.

Betslip extraction

Identical contract: send a screenshot with Read this betslip. and the response is one JSON object in the canonical tree — bets[] → components[] → selections[]. The full field-by-field schema and a worked NFL/NBA example live in the Sports-1 schema reference, and Sports-2 answers to every field of it.

On SharpBench Betslip (90 real slips, schema v3.1) Sports-1 measures 95.9% field-level accuracy and Sports-2 91.9% — Sports-2 trails on the hardest shapes (exotics and system bets). A pipeline built against the schema can switch between the two with one string change.

Pricing

Current per-token pricing is on the model page.