🤖 AI benchmark: hit-rate of 7 models

Prematch Ao vivo (in-play)

Seven external AI models (Hermes contour) independently analyze the same jogos — predicting the resultado (1X2), total (Mais de/Menos de), both equipes to placar (BTTS) and the exact placar. Here we honestly compare their palpites against the real result after the final whistle and combine everything into a single accuracy rating. An informational and analytical snapshot, not betting advice.

⚠️ Data is still accumulating — counting starts from 09.07.2026, so all models are compared on the same events (early test palpites are excluded). The sample is still small and not representative. Right now the snapshot holds 2524 jogo(es), 8699 settled AI palpites (Football, Tennis, Basketball, Volleyball, Hockey, Tabela tennis). The figures below are N, not «a percentage you can trust»: the more jogos are played out, the more reliable the snapshot becomes. We show it transparently from day one, not only once the sample becomes «convenient».

Leaderboard

Model N (settled) 1X2 Dupla chance (1X) Exact placar Composite accuracy
ChatGPT
263 65.0%(171/263) 74.6%(47/63) 17.2%(35/203) 44.2%(206/466)
Claude
1996 61.4%(1225/1996) 77.8%(543/698) 21.9%(376/1717) 43.1%(1601/3713)
Kimi
1109 60.8%(674/1109) 73.8%(253/343) 20.3%(196/964) 42.0%(870/2073)
Google AI
2634 60.3%(1588/2632) 74.5%(663/890) 20.4%(528/2592) 40.5%(2116/5224)
GLM 5.2
1285 59.5%(765/1285) 74.4%(302/406) 19.5%(246/1261) 39.7%(1011/2546)
DeepSeek
175 60.6%(106/175) 56.1%(23/41) 18.5%(32/173) 39.7%(138/348)
Qwen
1237 59.3%(733/1237) 75.0%(267/356) 16.8%(196/1166) 38.7%(929/2403)

grey — sample <5, not representative; «—» — the model has not made a settled palpite yet.

Dupla chance (1X) — the same escolha counts as a vitória if the chosen side won or the jogo drew. Of the 1X2 losses in football/hockey: 646 draws, 699 underdog (total settled 1X2 escolhas in these sports: 2797, double chance combined 75.0% (2098/2797)). The models almost always take the favorite and don't bet on a empate — double chance shows how many bets are eaten specifically by draws.

Composite accuracy — the share of correct palpites em all exibido mercados together: (sum of correct escolhas) ÷ (sum of all settled escolhas) em the mercados 1X2 + Exact placar. Each mercado-escolha weighs equally. This is hit-rate, not profitability — for money/ROI by model see /ai-agent. «Exact placar» — the full final placar was guessed correctly (H and A matched); for tennis, sets are compared; palpites with no recognized placar do not count toward the denominator. Dupla chance (1X): a escolha counts as a vitória if the chosen side won OR it was a empate — it accounts for frequent draws that «eat» bets on the favorite. This metric is informational and is not included in composite accuracy. On «Todos os esportes» we don't show total and BTTS — they're tied to the sport and make no sense in a mixed pool.

Composite model rating · all mercados

Bar height = the model's composite accuracy em all applicable mercados on the current sample. Sorted from best to worst.

44.2% (206/466)
GPT 5.5
43.1% (1601/3713)
Opus 4.8
42.0% (870/2073)
Kimi 2.6
40.5% (2116/5224)
Gemini 3.5 Flash
39.7% (1011/2546)
GLM 5.2
39.7% (138/348)
DeepSeek V4 Pro
38.7% (929/2403)
Qwen 3.7 Plus

Bars are AI models by version; grey/dimmed — sample <5, not representative. The snapshot is informational, not betting advice.

Accuracy by mercado

Where each model is strong: one mini-bar per applicable mercado, with the percentage and (hits/sample).

ChatGPT Composite 44.2%
1X2
65.0% (171/263)
Exact placar
17.2% (35/203)
Claude Composite 43.1%
1X2
61.4% (1225/1996)
Exact placar
21.9% (376/1717)
Kimi Composite 42.0%
1X2
60.8% (674/1109)
Exact placar
20.3% (196/964)
Google AI Composite 40.5%
1X2
60.3% (1588/2632)
Exact placar
20.4% (528/2592)
GLM 5.2 Composite 39.7%
1X2
59.5% (765/1285)
Exact placar
19.5% (246/1261)
DeepSeek Composite 39.7%
1X2
60.6% (106/175)
Exact placar
18.5% (32/173)
Qwen Composite 38.7%
1X2
59.3% (733/1237)
Exact placar
16.8% (196/1166)

The model's favorite by 1X2 = the max of P1/X/P2 in its probabilities; for sports without a empate (tennis, volleyball, etc.) the «X» option doesn't participate. grey — sample <5, not representative. Not betting advice.