Results
Hockey results Tuesday 20 October 2026: AI predictions review
Hockey results Tuesday 20 October 2026: the AI consensus went 3 of 6 across six NHL games, with two unanimous calls among the misses. Full review.
Six NHL games, ten frontier models and a single question before each puck drop: who wins? The answer, graded against the final scores, was a split verdict. The consensus finished 3 of 6 on Tuesday 20 October 2026, a night when the models read the tight games well and misread the lopsided ones badly.
The biggest surprise came in Washington, where all ten models backed the Capitals and all ten watched the Los Angeles Kings skate away with a 5-2 win. That was one of two unanimous picks on the card, and both of them went down. The other was in New Jersey, where the Colorado Avalanche were expected to edge a high-scoring night and instead lost 4-3.
The ice after the final buzzer on a night the consensus finished 3 of 6.
Where the AI got it right
The Utah Mammoth vs Pittsburgh Penguins prediction landed. Seven of ten models backed Utah, the most common scoreline was 4-3 and the Mammoth delivered a 5-4 win at home. The margin was a goal wider than the consensus expected, but the direction and the one-goal character of the game were read correctly at 57 per cent average confidence.
The New York Islanders vs Anaheim Ducks prediction was the cleanest call of the night. Nine of ten models took the Islanders, the most common score was 3-2, and the Islanders won 3-2 exactly. Average confidence sat at 57 per cent, but the degree of agreement was as strong as the desk saw all evening.
The Columbus Blue Jackets vs Toronto Maple Leafs prediction also came in. Seven of ten models favoured Toronto, the most common score was 4-5 and the Maple Leafs won 6-2. The winner was right; the expected margin was not, with Toronto turning a projected one-goal game into a four-goal one. That was the only correct call of the night that came with a comfortable final score.
Where it got it wrong
Washington was the night's clearest miss. Ten of ten models backed the Capitals, the most common score was 3-2, and average confidence was 58 per cent. The Kings won 5-2. There was no ambiguity in the consensus and no ambiguity in the result: a unanimous pick, beaten by three goals.
New Jersey was the second unanimous failure. All ten models expected the Colorado Avalanche to win, with the most common score 4-5 and average confidence at 58 per cent. The Devils won 4-3, so the models had the goals roughly right and the winner entirely wrong. Two 100 per cent consensus picks, two defeats.
Montreal was a softer miss but a miss nonetheless. Eight of ten models expected the Canadiens to win, the most common score was 4-3 and confidence averaged 57 per cent. The San Jose Sharks won 4-3, which means the models had the exact scoreline in mind and simply assigned it to the wrong team.
Models of the day
Five models tied for the best tally on the day. GPT-5.6 Luna, DeepSeek V4 Pro, Gemini 3.8 Flash, Kimi K3 and GLM 5.3 each finished 3 of 6 correct, matching the consensus record exactly. None of them found a way past the Washington and New Jersey results, but none of them fell below the group either.
The other five models finished 2 of 6. Claude Sonnet 5, Mistral Medium 3.5, Qwen 3.8 Max, MiMo V2.5 Pro and Grok 4.6 each dropped one call the leading group held. On a night with two unanimous misses, that single extra error was the whole difference between the top and bottom of the table.
Full results
- NHL: Utah Mammoth 5-4 Pittsburgh Penguins — consensus Utah Mammoth win (70%) RIGHT
- NHL: New York Islanders 3-2 Anaheim Ducks — consensus New York Islanders win (90%) RIGHT
- NHL: Washington Capitals 2-5 Los Angeles Kings — consensus Washington Capitals win (100%) WRONG
- NHL: New Jersey Devils 4-3 Colorado Avalanche — consensus Colorado Avalanche win (100%) WRONG
- NHL: Montreal Canadiens 3-4 San Jose Sharks — consensus Montreal Canadiens win (80%) WRONG
- NHL: Columbus Blue Jackets 2-6 Toronto Maple Leafs — consensus Toronto Maple Leafs win (70%) RIGHT
The pattern is worth noting for the next slate: the models were reliable when the panel was split and unreliable when it was not. Both unanimous calls failed, and the one near-unanimous call produced the most precise result of the night. Today's predictions are already live at TuringStats hockey predictions, where the panel will try to turn 3 of 6 into something better.
Next up: today's hockey predictions from the same panel, and the results archive with every graded call. Forecasts are informational only.
