NRL 2026 Season Review: How Our Tips Actually Went
96 from 156 published tips correct, 61.5%. Every tip graded exactly as published, what the confidence labels were worth, and what changes for 2027.
Quick answer. Across the 2026 NRL season our published tips went 96 from 156, or 61.5%. That counts every pick exactly as it appeared on TippingEdge before kick-off, from Round 8, when the site launched, to the grand final. It was 62.6% in the regular season and 4 from 9 in the finals.
A tipping site that only reports its good weeks is not worth reading. This is the whole season, graded the strict way: the pick that was on the page when the game kicked off, against the final score. Nothing re-run, nothing adjusted after the fact.
How we counted
- Picks as published. Each pick is taken from the round post as it appeared on the site. We checked the publish time of every post against every kick-off; all 156 tips went up before their game started.
- Rounds 8 to 31. The site launched during Round 8, so Rounds 1 to 7 are not counted. Round 8's post went up after that round's first game, so it only carried tips for the seven games still to come.
- Results are from the official NRL draw. There were no draws to leave out.
Did the confidence labels mean anything?
Every tip carries a confidence label based on the model's probability. If the model is well calibrated, a tip it rates at 70% should come in about 70% of the time.
| Label | Tips | Correct | Hit rate | Model's average stated chance |
|---|---|---|---|---|
| High | 48 | 36 | 75% | 81% |
| Medium | 50 | 29 | 58% | 70% |
| Low | 44 | 23 | 52% | 59% |
| Coin Flip | 14 | 8 | 57% | 53% |
The honest reading: the model was too sure of itself in every band except coin flips. High-confidence tips were the most reliable group, 36 from 48 (75%), but the model had rated them at 81% on average. The gap was widest for medium tips: rated around 70%, they landed 58% of the time. Overall the model's average stated chance was 68.6% against a real hit rate of 61.5%, about 7 points of overconfidence. That is the most useful thing this season taught us, and the first thing we are fixing for 2027.
The finals
4 from 9. The model's weakest month, as we warned it might be: finals put the strongest sides against each other and squeeze the gap a form-based model relies on. It did pick the premiers, the Roosters at 62% in a grand final they won by one point. Every finals tip is on the 2026 finals page.
Round by round
| Round | Tips correct |
|---|---|
| Round 8 | 7 of 7 |
| Round 9 | 2 of 8 |
| Round 10 | 6 of 8 |
| Round 11 | 6 of 8 |
| Round 12 | 3 of 5 |
| Round 13 | 4 of 7 |
| Round 14 | 6 of 8 |
| Round 15 | 4 of 5 |
| Round 16 | 4 of 7 |
| Round 17 | 6 of 8 |
| Round 18 | 3 of 5 |
| Round 19 | 4 of 7 |
| Round 20 | 5 of 8 |
| Round 21 | 6 of 8 |
| Round 22 | 5 of 8 |
| Round 23 | 4 of 8 |
| Round 24 | 3 of 8 |
| Round 25 | 4 of 8 |
| Round 26 | 6 of 8 |
| Round 27 | 4 of 8 |
| Round 28 (finals) | 2 of 4 |
| Round 29 (finals) | 0 of 2 |
| Round 30 (finals) | 1 of 2 |
| Round 31 (finals) | 1 of 1 |
Every individual tip, with the score, is on the results tracker.
Why our accuracy figure changed
Until October our accuracy tracker showed a higher number, 74.5% by the end of the season. It was calculated differently: it re-ran the current model over past rounds after that model had already been trained on those same games. That is a fair way to check the code works and an unfair way to measure the tips, because the model had effectively seen the answers.
The published record is what you would actually have got following the tips, and from 7 October 2026 it is the only accuracy figure we quote for the NRL model: on the tracker, in every round post and on the homepage. It also lines up with how the model does on seasons it was never trained on, around 62 to 63%.
What changes for 2027
- Calibration. The model's probabilities get re-calibrated so a 70% tip means 70%, starting with the medium band where the gap was widest.
- Data fixes. We found and fixed two errors in our historical match data: some grand finals were stored more than once and others were missing. Both are corrected.
- One accuracy number. Done: the published record, counted the way this page counts it. Each round's tips are now locked in the moment they are published.
Related on TippingEdge
Frequently asked questions
How accurate were TippingEdge's NRL tips in 2026?
Our published tips went 96 from 156, or 61.5%, from Round 8 to the grand final, counting every pick exactly as published before kick-off. That was 62.6% in the regular season and 4 from 9 in the finals.
Did TippingEdge tip the 2026 NRL premiers?
Yes. The model picked the Roosters at 62% for the grand final, and they beat the Knights 19-18.
Were the high-confidence tips reliable?
They were the most reliable group, 36 from 48 (75%), though still below the 81% the model claimed for them on average. Medium tips had the widest gap. Fixing that overconfidence is the main change for 2027.
Why did the TippingEdge accuracy figure drop in October 2026?
The old tracker re-ran the current model over past rounds after it had been trained on those games, which overstated the record at 74.5%. From 7 October 2026 we quote only the record of tips as actually published before kick-off, which was 61.5% for the season.