Reproducible, no-key-required benchmarks of sports odds APIs: ParlayAPI, The Odds API, OddsPapi, SharpAPI, SportsGameOdds, TheRundown.
Disclosure: this repo is maintained by ParlayAPI's founder. That is exactly why every number here is produced by a script you can run yourself, from this repo alone, with no API keys and no accounts. If a cell is wrong, the probe or the citation is wrong, and a PR fixes it. PRs from competitors are welcome and will be merged if honest.
Latest run: 2026-08-29, committed under results/2026-08-29/. A GitHub Action re-runs the probes weekly (Monday 06:00 UTC). See METHODOLOGY.md for scope and limitations, which are significant: v1 measures only what is measurable without paid keys. It says nothing about paid-tier data quality, odds accuracy, or update speed.
git clone https://github.com/JacobiusMakes/odds-api-benchmarks
cd odds-api-benchmarks
python3 bench/run_all.py # stdlib only; uv run bench/run_all.py also worksThat reruns every probe from your machine and regenerates this README from your results.
Single GET to each vendor's documented endpoints with no API key. Raw outcomes in results/2026-08-29/unauth.json.
| Vendor | Keyless endpoints found | Probe outcomes |
|---|---|---|
| ParlayAPI | 6 of 6 probed | 200 /v1/sandbox/sports; 200 /live/api/sports; 200 /v1/meta/credit-costs; 200 /v1/meta/source-quality; 200 /v1/calc/kelly; 200 /v1/prediction-markets/search |
| The Odds API | 0 of 1 probed | 401 /v4/sports |
| OddsPapi | 0 of 1 probed | 401 /v4/sports |
| SharpAPI | 0 of 1 probed | 401 /api/v1/odds |
| SportsGameOdds | 0 of 1 probed | 401 /v2/sports |
| TheRundown | 2 of 2 probed | 200 /api/v2/sports; 200 /api/v2/affiliates |
Probe targets and the docs page documenting each are listed in data/vendors.json. If a vendor has a keyless endpoint this benchmark missed, PR the URL and it will be probed next run.
These are the vendors' own claims from their own public pages, quoted with source URL and read date. They are not measurements. Snapshot: results/2026-08-29/free_tier.json.
| Vendor | Free allowance | Card required | Free tier delay | Book count claim | Source |
|---|---|---|---|---|---|
| ParlayAPI | 1,000 credits/month per the pricing page (the homepage phrases it as 1,000 requests/month; per-endpoint credit costs are published at /v1/meta/credit-costs) | No ('no credit card required') | Not stated as a delay; free tier includes 48 hours of historical access | '30+ sportsbooks' on both the homepage and the pricing page | parlay-api.com/pricing (read 2026-08-29), parlay-api.com (read 2026-08-29) |
| The Odds API | 500 credits/month (Starter plan) | Not stated on the cited page | Not stated on the cited page | 'over 40 bookmakers' globally | the-odds-api.com (read 2026-08-27) |
| OddsPapi | Not published as a fixed number on the pricing page (the page is a build-your-own-plan calculator); a vendor blog post describes a free tier of 250 requests | 'No credit card required' per vendor blog | Not stated on the cited pages | '350+ bookmakers' (homepage claims 300+ in one section and 350+ elsewhere; both numbers appear on vendor pages) | oddspapi.io/us/pricing (read 2026-08-27), oddspapi.io (read 2026-08-27), oddspapi.io/blog/free-odds-api-350-bookmakers (read 2026-08-27) |
| SharpAPI | 12 requests/min (~17,280/day), no monthly cap stated | No ('no credit card') | 60s data delay on free tier; pre-match only (live polling requires Pro+) | 2 sportsbooks on free tier (DraftKings, FanDuel); '40+ US sportsbooks' overall | sharpapi.io/pricing (read 2026-08-27) |
| SportsGameOdds | 2,500 objects/month (Amateur plan), 10 requests/min | Not stated on the cited page | 10-minute update frequency on free tier | 9 bookmakers on free tier; 82+ on top plans | sportsgameodds.com/pricing (read 2026-08-27) |
| TheRundown | 20,000 data points/day and 200,000/month, 1 req/sec | No ('No credit card') | 5-minute delay on free tier | '20+ sportsbooks and exchanges' including Kalshi, Polymarket, Novig, ProphetX; free tier includes 3 sportsbooks | therundown.io/pricing/api (read 2026-08-27), docs.therundown.io/quickstart (read 2026-08-27) |
Raw outcomes in results/2026-08-29/docs.json. "no" means not found at the candidate URLs in data/vendors.json; PR the correct URL if one exists elsewhere.
| Vendor | Machine-readable OpenAPI spec | llms.txt |
|---|---|---|
| ParlayAPI | yes (spec) | yes (llms.txt) |
| The Odds API | not found at probed URLs | not found at probed URLs |
| OddsPapi | not found at probed URLs | yes (llms.txt) |
| SharpAPI | yes (spec) | yes (llms.txt) |
| SportsGameOdds | yes (spec) | yes (llms.txt) |
| TheRundown | not found at probed URLs | yes (llms.txt) |
Raw outcomes in results/2026-08-29/transparency.json. History columns are shallow keyword checks against the fetched HTML; JS-rendered status pages can under-report, so treat a "no" here as "not detected", not "absent".
| Vendor | Public status page | Uptime history detected | Incident/update history detected |
|---|---|---|---|
| ParlayAPI | yes (link) | no | no |
| The Odds API | yes (link) | yes | yes |
| OddsPapi | not found at probed URLs | n/a | n/a |
| SharpAPI | yes (link) | yes | yes |
| SportsGameOdds | not found at probed URLs | n/a | n/a |
| TheRundown | not found at probed URLs | n/a | n/a |
n=20 sequential GETs per target, median and p90 of total round-trip time. Raw samples context in results/2026-08-29/latency.json. A p50/p95 summary table with success rates is regenerated into RESULTS.md on every run, along with the badge data in badge-latency.json.
Caveats, do not skip: Single vantage point. Includes TLS + full response for small payloads. 401 responses measure auth-rejection latency, not data latency. Not comparable to odds update frequency.
| Vendor | Target | Statuses seen | Median | p90 |
|---|---|---|---|---|
| ParlayAPI | keyless API endpoint /live/api/sports | [200] | 143 ms | 723 ms |
| The Odds API | API edge: /v4/sports (returns 401 without key; measures edge round trip only) | [401] | 63 ms | 164 ms |
| OddsPapi | API edge: /v4/sports (returns 401 without key; measures edge round trip only) | [401] | 124 ms | 134 ms |
| SharpAPI | API edge: /api/v1/odds (returns 401 without key; measures edge round trip only) | [401, 429] | 26 ms | 46 ms |
| SportsGameOdds | API edge: /v2/sports (returns 401 without key; measures edge round trip only) | [401] | 71 ms | 89 ms |
| TheRundown | keyless API endpoint /api/v2/sports | [200] | 40 ms | 53 ms |
A benchmark maintained by one vendor's founder owes you this section. Based on the latest committed results and the cited vendor pages:
- The Odds API runs a public status page with visible uptime history and status update history, which ParlayAPI's live status page does not show (ParlayAPI's page shows current per-source freshness, not historical uptime). Its free tier requires no configuration and its pricing page states plan sizes in one screen.
- SharpAPI offers a free tier of 12 requests/min with no stated monthly cap (roughly 17,280/day if sustained), publishes a public status page, an llms.txt, and a machine-readable OpenAPI spec that this benchmark's probes validated.
- TheRundown offers the largest stated daily free allowance in this comparison (20,000 data points/day per its pricing page), covers prediction-market exchanges (Kalshi, Polymarket, Novig, ProphetX), and its reference endpoints answered keyless probes with real JSON. It has also been the fastest keyless-200 endpoint in this repo's latency probe, faster than ParlayAPI's keyless endpoint in recent committed runs; see the latest latency table and RESULTS.md for current numbers.
- OddsPapi claims the largest bookmaker count in this comparison (350+ per its own pages, including sharp books and exchanges) and publishes an llms.txt. Its build-your-own pricing calculator lets you pay for only the books and sports you need.
- SportsGameOdds publishes both an llms.txt and a machine-readable OpenAPI spec that this benchmark's probes validated, and prices per event object rather than per market/book combination, which can be materially cheaper for wide market coverage; its pricing page states plan terms plainly.
- ParlayAPI (the maintainer's product) answered the most keyless probes in this run, but the probe list is curated by the maintainer (competitor keyless endpoints are added by PR), and note what it lacks versus the others: no public uptime/incident history was detected on its status page, its headline book count (30+ sportsbooks per its own pages) is smaller than most competitors' headline counts in this comparison, and its keyless endpoint has not been the fastest in this repo's own latency probe.
If any of this is stale or wrong, open a PR with the vendor's current page and read date.
PRs welcome, including and especially from the vendors benchmarked here. Ground rules: every claim needs either a probe script that reproduces it or a citation to a vendor's own page with URL and read date. Fixes to probe targets, added vendors, and corrections to stale free-tier terms are all fair game. Marketing copy is not.
MIT. Results data is factual output of the probes and carries no warranty.