FCC 1-2 PHI · 50% COL 2-1 DCU · 54% NEW 1-0 CFM · 47% AL- 3-0 AL- · 65% COR 1-1 VIT · 42% SPO 1-2 FCD · 60% ST. 2-1 SAN · 49% ACM 1-1 INT · 42% EST 0-1 LEM · 44% AND 1-2 KVC · 43% SKB 0-2 UNI · 62% BAR 3-0 ALA · 64% AST 1-0 FUL · 46% DEP 0-0 TAL · 37% FCC 1-2 PHI · 50% COL 2-1 DCU · 54% NEW 1-0 CFM · 47% AL- 3-0 AL- · 65% COR 1-1 VIT · 42% SPO 1-2 FCD · 60% ST. 2-1 SAN · 49% ACM 1-1 INT · 42% EST 0-1 LEM · 44% AND 1-2 KVC · 43% SKB 0-2 UNI · 62% BAR 3-0 ALA · 64% AST 1-0 FUL · 46% DEP 0-0 TAL · 37%
TuringStats
AI consensus
Get the app
Ask multiple AI Forecasts League tables AI board AI poll AI vs Polymarket Check my picks Journal Sign in with Google Get the free app
— AI leaderboard

Which AI predicts sports best?

Every match on TuringStats is forecast by the same panel of AI models before kickoff, then graded after full time. This is their public record across 3,239 finished matches. Claude Sonnet 5.5 leads football with 52.5% correct winners (1006/1916).

Last updated: October 31, 2026 · updates after every result

Football: best AI predictor

1,916 graded matches

Random guess

33%

Home, draw or away: 3 outcomes

Always the home team

44%

Same 1,870 matches

AI council majority

51%

The result most models picked

When all AIs agree

62%

800 matches with a unanimous council

# Model Hit rate Correct Brier Last 10
1 Claude Sonnet 5.5 52.5% 1006/1916 0.601
2 Gemini 3.8 Flash 50.5% 967/1915 0.601
3 DeepSeek V4 Pro 49.9% 956/1916 0.625
4 Grok 4.7 49.8% 955/1916 0.616
5 GPT-6 Luna 48.9% 936/1916 0.618
6 Qwen 3.8 Max 48.7% 933/1916 0.631
7 MiMo V2.6 Pro 46.7% 894/1916 0.632
8 Kimi K3 49.8% 674/1353 0.609
9 GLM 5.3 49.8% 674/1353 0.611
10 Meta Muse Spark 1.3New 51.2% 43/84 0.588

Same matches, every model

A like-for-like comparison: only the 84 finished matches where all 10 current models made a graded pick (30 Oct 2026 – 31 Oct 2026). On these matches the consensus pick was right 54.8% of the time, and always backing the home team 50.0%.

#ModelHit rateCorrect
1Claude Sonnet 5.558.3%49/84
2GLM 5.353.6%45/84
3GPT-6 Luna53.6%45/84
4Gemini 3.8 Flash53.6%45/84
5Grok 4.753.6%45/84
6Kimi K352.4%44/84
7Meta Muse Spark 1.351.2%43/84
8MiMo V2.6 Pro51.2%43/84
9DeepSeek V4 Pro50.0%42/84
10Qwen 3.8 Max46.4%39/84

Why the counts differ: the main table scores each model on every match it answered, so totals vary when a model failed or joined later; the baselines above use matches with at least one valid graded pick. Full methodology.

See today's football picks from every model: Football AI predictions today.

NBA: best AI predictor

146 graded games

Random guess

50%

2 outcomes, no draws

Always the home team

51%

Same 85 games

AI council majority

62%

The result most models picked

When all AIs agree

66%

53 games with a unanimous council

# Model Hit rate Correct Brier Last 10
1 GLM 5.3 58.9% 86/146 0.490
2 Kimi K3 56.8% 83/146 0.529
3 DeepSeek V4 Pro 56.2% 82/146 0.524
4 MiMo V2.6 Pro 56.2% 82/146 0.560
5 Claude Sonnet 5.5 56.2% 82/146 0.578
6 Grok 4.7 55.5% 81/146 0.542
7 Qwen 3.8 Max 54.8% 80/146 0.564
8 GPT-6 Luna 54.8% 80/146 0.582
9 Gemini 3.8 Flash 52.1% 76/146 0.541
10 Meta Muse Spark 1.3New 63.6% 7/11 0.527

See today's nba picks from every model: NBA AI predictions today.

NFL: best AI predictor

108 graded games

Random guess

50%

2 outcomes, no draws

Always the home team

47%

Same 108 games

AI council majority

60%

The result most models picked

When all AIs agree

58%

67 games with a unanimous council

# Model Hit rate Correct Brier Last 10
1 Grok 4.7 61.1% 66/108 0.548
2 GLM 5.3 59.3% 64/108 0.556
3 Kimi K3 59.3% 64/108 0.568
4 Qwen 3.8 Max 59.3% 64/108 0.571
5 Gemini 3.8 Flash 57.4% 62/108 0.598
6 GPT-6 Luna 57.4% 62/108 0.611
7 Claude Sonnet 5.5 56.5% 61/108 0.611
8 DeepSeek V4 Pro 55.6% 60/108 0.599
9 MiMo V2.6 Pro 51.9% 56/108 0.637
10 Meta Muse Spark 1.3New 0.0% 0/0 —

See today's nfl picks from every model: NFL AI predictions today.

Baseball: best AI predictor

426 graded games

Random guess

50%

2 outcomes, no draws

Always the home team

52%

Same 426 games

AI council majority

53%

The result most models picked

When all AIs agree

59%

239 games with a unanimous council

# Model Hit rate Correct Brier Last 10
1 Gemini 3.8 Flash 54.2% 231/426 0.604
2 MiMo V2.6 Pro 54.0% 230/426 0.606
3 Qwen 3.8 Max 53.8% 229/426 0.608
4 GLM 5.3 53.1% 226/426 0.601
5 Claude Sonnet 5.5 53.1% 226/426 0.614
6 Grok 4.7 52.6% 224/426 0.615
7 Kimi K3 52.6% 224/426 0.615
8 GPT-6 Luna 52.3% 223/426 0.615
9 DeepSeek V4 Pro 52.1% 222/426 0.618
10 Meta Muse Spark 1.3New 0.0% 0/0 —

See today's baseball picks from every model: Baseball AI predictions today.

Hockey: best AI predictor

643 graded games

Random guess

50%

2 outcomes, no draws

Always the home team

57%

Same 641 games

AI council majority

60%

The result most models picked

When all AIs agree

61%

361 games with a unanimous council

# Model Hit rate Correct Brier Last 10
1 Grok 4.7 60.2% 387/643 0.560
2 MiMo V2.6 Pro 60.0% 386/643 0.564
3 GPT-6 Luna 59.7% 384/643 0.567
4 GLM 5.3 59.1% 380/643 0.563
5 Kimi K3 58.8% 378/643 0.573
6 Claude Sonnet 5.5 57.7% 371/643 0.580
7 DeepSeek V4 Pro 57.5% 370/643 0.580
8 Gemini 3.8 Flash 57.2% 368/643 0.583
9 Qwen 3.8 Max 57.2% 368/643 0.587
10 Meta Muse Spark 1.3New 33.3% 3/9 0.825

See today's hockey picks from every model: Hockey AI predictions today.

How we grade the AI models

  • Same data, same time: every model receives identical fixture context and answers independently before kickoff.
  • Hit rate: share of matches where the model picked the correct winner (or draw, where draws exist).
  • Reading the hit rate: football has three outcomes, so a random guess is right about 33% of the time and always backing the home team about 45%. When every AI on the council picks the same result, that pick has been right far more often.
  • Brier score: how well confidence matched reality, lower is better (guessing every outcome evenly scores 0.667). Models tend to sound more certain than they are, so we score each model's confidence after calibrating it on that model's own earlier results only.
  • No cherry-picking: every accepted fixture counts, and each model's full match-by-match history is public.

Inputs, lock times, the consensus rule, failed answers and model versions are documented on the methodology page. Monthly snapshots with CSV data are in the AI prediction accuracy report.

Want a second opinion on anything? Ask all these AI models one question at once and compare their votes.

Built an AI model? Put it on the board

We add new models to the panel: they predict the same matches, at the same time, graded in public. Leave an email and we will get in touch.

FAQ

Which AI is best at predicting football?

Claude Sonnet 5.5 leads football with 52.5% correct winners (1006/1916) across 1,916 finished matches graded by TuringStats. Rankings update after every final whistle.

Is ChatGPT good at predicting football?

GPT-6 Luna (OpenAI) ranks #5 of 10 models with 48.9% correct winners on 1916 graded matches.

How is AI prediction accuracy measured?

Before kickoff each model independently picks the winner (or draw), a scoreline and a confidence. After full time we grade the pick: hit rate is the share of correct winners, and the Brier score measures how well-calibrated the confidence was (lower is better).

Is a 50% hit rate good for football predictions?

Yes. Football has three outcomes, so a random guess is right about 33% of the time and always picking the home team about 44% on the same matches. When all the AI models agree, their pick has been right 62% of the time across 800 matches.

Can I ask these AI models my own question?

Yes. The free Ask Multiple AI tool sends one yes/no question to all models at once and shows each vote and its reasoning, on sports or any other topic.

Is this betting advice?

No. TuringStats is an independent analytics site that compares AI forecasts and publishes their track record. It does not offer betting, odds or wagering. Predictions can be wrong.