The AI Lab · Silicon Tipster League
DeepSeek: horse racing tips, tested
DeepSeek · DeepSeekCurrently #4 of 8
DeepSeek is the lean outsider in our five-AI league, the one that made its name doing more with less. Here you can watch it read each racecard cold and commit to a single horse, in public, with nothing hidden.
Research, not tips. Every pick logged before the off and settled at industry Starting Price, wins and losses alike. 18+ · please gamble responsibly.
DeepSeek's scorecard
2,328 picks settledStaked (£1 stakes)
£2,328
Returned
£1,938
Profit
−£390
ROI
-16.7%
AI Picks the Winner
−£245(-18.5%)Never sees the odds, reads the race itself.
- Staked
- £1,326
- Returned
- £1,081
- Bets
- 1,326
- Win%
- 22%
AI Finds the Value Bets
−£144(-14.4%)Sees the market's implied chance, hunts the value it missed.
- Staked
- £1,002
- Returned
- £858
- Bets
- 1,002
- Win%
- 16%
Went its own way: 19 races where DeepSeek was the lone dissenter while the rest of the field agreed on a different horse.
DeepSeek's running profit: AI Picks the Winner vs AI Finds the Value Bets
One line per arm, £1 a bet, settled to Starting Price. The gap between them is the interesting bit: does seeing the market's view help DeepSeek, or just pull it onto the favourite?
DeepSeek's recent calls
The actual picks and one-line reasons DeepSeek logged before each off, newest first. Read the logic and judge it for yourself. The board settles the argument at Starting Price.
| Race | Arm | Pick | Conf. | Reason | Result |
|---|---|---|---|---|---|
| Kempton Park21:00 | Value pick | Piazza | Medium | Course form and trip suit; market underestimates. | Won · SP 10 · +£9.00 |
| Kempton Park21:00 | Winner pick | Viviana | Medium | Loughnane aboard, consistent form, handles 7f well. | Lost |
| Kempton Park20:30 | Value pick | White Crown Star | Low | Market underestimates; trainer form and trip suit. | Lost |
| Kempton Park20:30 | Winner pick | Rogue Defence | Medium | Loughnane aboard, strong recent form, suitable conditions. | Lost |
| Sligo20:18 | Value pick | Orderly Conduct | Medium | Cromwell runner overpriced; market underestimates each-way claims. | Lost |
| Sligo20:18 | Winner pick | Run For Fun | High | Top Mullins stable, strong bumper form, likely class edge. | Lost |
| Worcester20:08 | Value pick | Farhh Echo | Low | Huge price, trip suits, trainer form unknown. | Lost |
| Worcester20:08 | Winner pick | Mancero | Medium | Strong stable form, trip suits, Bowen aboard. | Lost |
| Kempton Park20:00 | Value pick | Exposure | Medium | Gosden yard, strong form, market underestimates. | Lost |
| Kempton Park20:00 | Winner pick | Exposure | Medium | Gosden yard strong, recent form solid, conditions suit. | Lost |
| Sligo19:48 | Value pick | Strong Contender | Medium | Market overrates favourite; this one has solid form and conditions suit. | Won · SP 5 · +£4.00 |
| Sligo19:48 | Winner pick | Strong Contender | Medium | Progressive profile, good ground suits, trainer in form. | Won · SP 5 · +£4.00 |
About DeepSeek
DeepSeek is the AI built by the Chinese lab of the same name, and it arrived as the outsider that made the whole field sit up. Its calling card was cost-to-performance: matching the heavy hitters while spending a fraction of what the big Western labs pour into their models. That efficiency-first reputation is what put DeepSeek on the map almost overnight.
It is not the loudest name on the board and it does not need to be. The appeal is a model known for reasoning cleanly rather than throwing brute force at a problem. It reads the race, decides, and moves on. On a busy racecard, that stripped-back style is interesting to watch.
None of that reputation was built on horses, of course. A model that reasons efficiently about maths and code has never been asked to weigh a drop in trip against soft ground and a wide draw. That is exactly the gap the Lab exists to probe, and DeepSeek gets tested on the same racecards, under the same rules, as every rival.
How DeepSeek actually picks
Handed a racecard (runners, going, class, distance) DeepSeek has to turn that into one horse and one short line of reasoning, capped at roughly twelve words. What we will be watching is whether its efficient, no-frills style translates to the paddock: does a model that made its name by cutting to the essentials pick out the one factor that actually settles a race, or does the shorthand skate past the detail that horses live and die on? The reason strings are published alongside every call, so you can read exactly what it latched onto and judge for yourself whether the logic holds up once the stalls open.
Here is what the numbers say, updated live as the sample grows. Reading a race blind, with the prices hidden, DeepSeek backs the outright favourite 29%of the time, and its picks most resemble the “back short-priced favourites” approach (backs the favourite only when it is odds-on). Its stated reasons lean on form, trainer and draw.
Show it the market and it changes character: its value picks swing toward the “favourite in a big field” approach, hunting the overlooked rather than the well-fancied. Any positive return here is over a young sample and is not yet a proven edge, research, not tips.
See how all five AIs rank → · Read the full write-up on how DeepSeekpicks →
AI Picks the Winner vs AI Finds the Value Bets
DeepSeek runs in both tests at once. In AI Picks the Winner it never sees a price and must read the race on its own terms; in AI Finds the Value Bets it also sees the market's implied chance for each runner (never the SP). For a model whose reputation is built on doing more with less, AI Picks the Winner is the truer test: it is DeepSeek working from the form alone, with nothing to lean on. AI Finds the Value Bets is the more revealing one to watch: a market-shaped favourite is the easy, well-trodden answer, and the interesting question is whether an efficient reasoner treats the odds as one more input to weigh or quietly lets the price do its thinking for it. Watching the two tests diverge is where you learn how much of the pick is really DeepSeek and how much is the crowd.
What to watch on DeepSeek's board
- The Form-vs-Value split: does seeing the price sharpen DeepSeek's call, or just pull it toward the favourite everyone else backs?
- Its reason strings: whether that lean, cut-to-the-essentials style names the one factor that decides the race
- The races where DeepSeek went its own way, the lone dissenter while the other four AIs agreed on a different horse
- How a from-scratch board fills in live, race by race, with every call logged before the off
The rest of the field
DeepSeek is one of 8 on the board. See how the others are reading the same races:
Questions about DeepSeek tips
Is DeepSeek good at horse racing tips?
We do not know yet, and that is the point of running it live. DeepSeek earned its reputation on reasoning and efficiency, not on reading a racecard, so the Silicon Tipster League is a real test rather than a showcase. Every pick is logged before the off and settled honestly at Starting Price, so the board shows what actually happened, not a highlight reel.
Can I make money following DeepSeek's picks?
No, please do not treat these as a way to make money. Naming a likely winner is not the same as being paid enough when it wins, and the bookmaker's margin is built to beat exactly this kind of selection. The Lab's wider finding across tens of thousands of races is that no method tested clears that margin. Treat DeepSeek's calls as research and entertainment, nothing more.
Why does DeepSeek make two picks per race?
It runs in both tests side by side. In AI Picks the Winner it never sees the odds and must read the race itself; in AI Finds the Value Bets it also sees the market's implied chance for each runner, though never the SP. Running both lets us see how much the market moves DeepSeek's thinking, and whether it improves the call or just nudges it toward the crowd's favourite.
Is this betting advice?
No. This is an experiment and a piece of entertainment, not tips to bet on. It is intended for over-18s, none of it is financial or betting advice, and if you do choose to bet, only ever stake what you can afford to lose. Please gamble responsibly.
This page sits inside the AI Lab, where we test whether any betting system makes money (across 26,000+ races, none of them do), and ask the bigger question in does following an AI tipster work?
Gamble responsibly.This page is research and entertainment, not betting advice. No AI here beats the bookmaker's margin, and nothing on it is a signal to stake. Betting should never be a way to make money. If it is affecting you or someone you know, free and confidential support is at BeGambleAware.org. 18+.