2026 season / Round 14reviewed

MADRING

Spanish Grand Prix

Compare what each AI predicted before the race with what happened.

Official podium

P1 Kimi Antonelli
P2 Max Verstappen
P3 Lando Norris

Full result ↗

After the race

How the models did

3/3 scored

Predictions scored against the official result. Lower error (RPS) is better; 0 is perfect. Retirement error is the average Brier score across all drivers, shown separately; it does not set the leaderboard order. Cost and time cover making the prediction.

Model	Error	Retirement error	Model API	Time	Rank error	Season error
Grok 4.6	0.0856	0.1395	$1.63	14.6m	2.83 places	0.1247
Muse Spark 1.3	0.0911	0.1387	$1.27	9.6m	3.05 places	0.1251
GPT-5.6 Sol	0.0959	0.1410	$2.70	20.2m	3.27 places	0.1330
Grid baseline	0.1157	—	—	—	—	—

Model API costs exclude historical search and extraction fees.

Grok 4.6 scored best. Its biggest gains over Muse Spark 1.3 came from Oliver Bearman, Franco Colapinto, Lance Stroll. Select a model below to inspect every driver’s contribution.

What is the grid baseline?

Predict every driver finishes in the official starting order, with 100% certainty and no retirements. Score it with the same RPS rule. This is a simple reference, not a competitive forecasting system.

Added retrospectively; the rule has no fitted parameters. Grids were retrieved on 26 September 2026, after these races. Late grid changes may differ from what was available when an AI ran. Pit-lane starters use their listed place in the official order.

Official starting grid ↗

Season average: 2 shared races in 2026. Rank error compares each driver’s expected finish with the result. Unclassified outcomes count as field size + 1.

Where the models disagree

Before-race chances, side by side. Actual result: P1.

Model	Pick	Win	Top 3	Retire chance	Retirement error
GPT-5.6 Sol	—	37.9%	73.8%	14.0%	0.0196
Muse Spark 1.3	—	37.1%	74.8%	12.4%	0.0153
Grok 4.6	—	36.5%	79.5%	10.0%	0.0100

Retirement error: lower is better; 0 is perfect. A higher retirement chance earns more credit when the driver retires, and less when they finish. This separate score does not set the leaderboard order.

— No explicit finishing-order pick was saved; the original probabilities are shown.

Forecast saved 13 Sept 2026, 07:37 UTCOriginal saved prediction · timing not recorded

Original prompt ↗Prediction JSON ↗Scoring code ↗

What drove GPT-5.6 Sol’s error?

Largest contributions: Lewis Hamilton (0.0292), Carlos Sainz (0.0092), Sergio Perez (0.0073).

These contributions add up to the race error of 0.0959. Lower is better. They measure the saved probabilities against the result, not the quality of the written explanation.

Driver	Result	Error added
Lewis Hamilton	NC · Retired	0.0292
Carlos Sainz	NC · Retired	0.0092
Sergio Perez	NC · Retired	0.0073
Max Verstappen	P2	0.0040
Franco Colapinto	P7	0.0038
Oscar Piastri	P8	0.0033
Oliver Bearman	P16	0.0029
Nico Hulkenberg	P10	0.0029
Gabriel Bortoleto	P13	0.0028
Charles Leclerc	P4	0.0028
Lance Stroll	NC · Retired	0.0027
Liam Lawson	P6	0.0027
Fernando Alonso	P17	0.0027
Lando Norris	P3	0.0025
Arvid Lindblad	P9	0.0025
Valtteri Bottas	P18	0.0023
Alexander Albon	P15	0.0023
Yuki Tsunoda	P14	0.0022
Kimi Antonelli	P1	0.0021
Pierre Gasly	P12	0.0021
Esteban Ocon	P11	0.0020
George Russell	P5	0.0018
Total (before rounding)		0.0959

NC = not classified; DNS = did not start; DSQ = disqualified. A retired driver can still have a classified position. All three unclassified outcomes sit after the last place for RPS. Scoring code ↗

Before the race

Forecast probabilities

22 drivers

Order by

Sorted by average rank across all outcomes. NC, DNS and DSQ count as rank 23; a larger downside risk can outweigh a higher podium chance. This archived run saved probabilities, not an explicit finishing-order pick. Select a driver for details.

Order	Win	Top 3	Top 10	Retire	Result
01	37.9%	73.8%	84.9%	14.0%	P1
02	34.5%	68.4%	81.4%	17.5%	P3
03	6.3%	39.6%	86.6%	12.0%	NC · Retired
04	4.3%	24.6%	83.2%	15.0%	P5
05	11.4%	54.3%	74.9%	24.0%	P2
06	3.0%	20.2%	78.1%	20.0%	P4
07	1.5%	9.7%	79.4%	18.0%	P8
08	0.5%	3.4%	77.6%	17.0%	P6
09	0.1%	1.2%	63.8%	14.5%	P7
10	0.1%	0.9%	47.6%	15.0%	P11
11	0.1%	0.9%	58.3%	20.0%	P9
12	0.1%	0.8%	40.4%	15.0%	P12
13	0.1%	0.7%	41.8%	15.0%	P13
14	0.1%	0.6%	45.6%	25.0%	P10
15	0.0%	0.4%	22.4%	17.0%	P14
16	0.0%	0.1%	9.1%	27.0%	P15
17	0.0%	0.1%	8.5%	19.0%	NC · Retired
18	0.0%	0.0%	3.5%	28.0%	NC · Retired
19	0.0%	0.0%	4.1%	35.0%	P17
20	0.0%	0.1%	5.3%	31.0%	P16
21	0.0%	0.0%	2.5%	36.0%	P18
22	0.0%	0.0%	1.1%	47.0%	NC · Retired

Retirement is a separate risk: a late retirement can still be classified. NC = not classified. Green only marks an exact finishing-position match; it is not a score. Retirement predictions are scored even without a green mark. Select a driver to see their retirement error.

Why it predicted this

GPT-5.6 Sol’s own explanation, written before the race: the evidence it used, its assumptions, and what could go wrong. Its claims are not independently verified.

Its favourite to win

Kimi Antonelli · 37.9% chance

Original research

3 sections · 505 words

2026 Spanish Grand Prix (MADRING) pre-race forecast

Forecast cut-off: 13 September 2026, approximately 07:30 UTC (race start 13:00 UTC).

Evidence used

The FIA provisional grid puts Norris, Antonelli, Verstappen, Hamilton, Leclerc, Russell, Piastri and Lawson in the first eight places. Sainz drops to P20 for impeding Bottas; Bearman and Stroll are permitted to start P21/P22; Stroll has a 40-place power-unit penalty. FIA Document 57 and permission-to-start decision.
Norris beat Antonelli by only 0.011 seconds in Q3, with Verstappen 0.140s from pole and the first seven covered by 0.470s. Earlier phases and all three practices pointed more strongly toward Mercedes than Norris's exceptional final lap alone. Official qualifying report.
Friday's short, traffic-affected long runs made Russell the benchmark. Motorsport's compound/stint-adjusted analysis had Verstappen about 0.13s per lap behind, Antonelli 0.55s behind, the Ferraris around 0.9s behind and Piastri 1.51s behind. Norris had no representative high-fuel run after a gearbox problem. Motorsport.com long-run analysis.
The strategic evidence is unusually uncertain. Pirelli measured roughly 0.2–0.3s/lap degradation and front-left graining, but the green-flag pit loss is about 24 seconds and overtaking was extremely difficult in the support races. Medium-Hard is the natural one-stop; Medium-Hard-Hard remains competitive if Friday's degradation persists. F1 team/Pirelli Saturday comments and strategy analysis.
The race is forecast to be hot, clear and dry, around 32°C ambient with a possible 53°C track temperature. Motorsport.com weather preview.
Post-qualifying market prices were used as an external calibration rather than as the model itself. At the cut-off, the liquid Polymarket mid-prices were approximately Antonelli 40%, Norris 35%, Verstappen 12%, Hamilton 6.5%, Russell 4.45%, Leclerc 2.9% and Piastri 1.7% before normalization; sportsbook prices showed a similar front-row concentration. Polymarket event and post-qualifying odds snapshot.

Method and interpretation

I combined the 13 completed 2026 Grands Prix, recent form, grid position, weekend pace and driver/team reliability in a Monte Carlo rank model. Because this is MADRING's first Grand Prix, outcomes were mixed across three latent race shapes: a track-position-led one-stop race, a high-degradation strategic race, and a Safety Car/red-flag-heavy race. Team-level pace uncertainty, starts, strategy errors, non-terminal incidents and correlated race conditions broaden the positional tails.

The resulting win probabilities are led by Antonelli (37.89%), Norris (34.51%), Verstappen (11.38%), Hamilton (6.27%), Russell (4.34%), Leclerc (3.03%) and Piastri (1.47%). Norris has the track-position advantage; Antonelli gets a narrow overall edge from Mercedes' high-fuel evidence and season-long strength.

Retirement risk is above a normal dry-race baseline because the circuit is wall-lined and unfamiliar, practice produced several heavy crashes, support racing produced a first-lap pile-up, and 2026 reliability has been poor. Stroll is the largest individual risk after repeated Honda failures and this weekend's fresh-PU water-pressure problem; Bearman also carries extra DNS/retirement risk while Haas rebuilds the car after his FP3 impact. The model averages about 4.82 race retirements.

retirement is modeled separately from classification. A driver retiring late can still receive a numbered FIA classification; earlier retirements go to nc. dns includes failure to take the start, while dsq is a small separate post-event risk.

Read the complete note

2026 Spanish Grand Prix (MADRING) pre-race forecast

Forecast cut-off: 13 September 2026, approximately 07:30 UTC (race start 13:00 UTC).

Evidence used

The FIA provisional grid puts Norris, Antonelli, Verstappen, Hamilton, Leclerc, Russell, Piastri and Lawson in the first eight places. Sainz drops to P20 for impeding Bottas; Bearman and Stroll are permitted to start P21/P22; Stroll has a 40-place power-unit penalty. FIA Document 57 and permission-to-start decision.
Norris beat Antonelli by only 0.011 seconds in Q3, with Verstappen 0.140s from pole and the first seven covered by 0.470s. Earlier phases and all three practices pointed more strongly toward Mercedes than Norris's exceptional final lap alone. Official qualifying report.
Friday's short, traffic-affected long runs made Russell the benchmark. Motorsport's compound/stint-adjusted analysis had Verstappen about 0.13s per lap behind, Antonelli 0.55s behind, the Ferraris around 0.9s behind and Piastri 1.51s behind. Norris had no representative high-fuel run after a gearbox problem. Motorsport.com long-run analysis.
The strategic evidence is unusually uncertain. Pirelli measured roughly 0.2–0.3s/lap degradation and front-left graining, but the green-flag pit loss is about 24 seconds and overtaking was extremely difficult in the support races. Medium-Hard is the natural one-stop; Medium-Hard-Hard remains competitive if Friday's degradation persists. F1 team/Pirelli Saturday comments and strategy analysis.
The race is forecast to be hot, clear and dry, around 32°C ambient with a possible 53°C track temperature. Motorsport.com weather preview.
Post-qualifying market prices were used as an external calibration rather than as the model itself. At the cut-off, the liquid Polymarket mid-prices were approximately Antonelli 40%, Norris 35%, Verstappen 12%, Hamilton 6.5%, Russell 4.45%, Leclerc 2.9% and Piastri 1.7% before normalization; sportsbook prices showed a similar front-row concentration. Polymarket event and post-qualifying odds snapshot.

Method and interpretation

After the race: its review

2 sections · 513 words

Written after seeing the result. These are the model’s own lessons, not proof that its next prediction will improve.

2026-14 Madrid post-race (official result)

Result: ANT-VER-NOR-LEC-RUS-LAW-COL-PIA-LIN-HUL; NC SAI/PER/STR/HAM. ANT was the 37.9% P1 favourite, the three highest podium probabilities made the exact podium, and 9/10 drivers with the highest points probabilities scored (HUL, ranked 12th at 45.6%, replaced HAM). The central pace/order model was strong; do not materially retune it from this race.
Separate outcome from pace: NOR led by about 4.5s and had the best managed race pace. A Lap-13 VSC began just after he passed pit entry, gave ANT/VER cheap stops, and ended before NOR returned; his subsequent 7s stop compounded the loss. ANT winning is neutralisation/pit-stop variance, not proof that the sparse Friday long-run signal correctly put Mercedes ahead. Keep adjusted short practice runs heavily shrunk and treat a missing run as uncertainty, not negative evidence.
Friday graining/degradation did not translate directly: difficult passing, a ~24s stop loss, track evolution and managed pace made one stop dominant, while the little-used Hard stayed consistent for 45+ laps. At new tracks, condition multi-stop/high-degradation weight on passing incentives and likely management, and retain wider priors for untested compounds.
Model VSC/SC timing jointly with track location, starting compound and pit windows rather than only adding generic “chaos” dispersion; even a very short VSC produced large, correlated strategy gains/losses in an otherwise processional race.
Tail check, not a pace update: PIA's P8 had only 2.97% exact probability (largest log-loss miss) after Lap-1 front-wing damage, traffic and an adverse VSC. On low-overtaking tracks, ensure repairable contact carries enough classified multi-place downside mass rather than routing most adverse mass to retirement. HAM's brake DNF was a plausible 10.8% NC tail, not evidence to demote his pace.
Reliability level was sound: 4 retirements versus 4.82 expected (4.34 expected NC), retirement Brier 0.1410. Elevated STR/PER risks helped, but their realised causes do not cleanly validate every causal prior; do not retune individuals from one draw. Audit all post-crash rebuilds consistently by repair scope and successful subsequent mileage; Hamilton's brake root cause was unconfirmed.
Longitudinal score baseline: mean RPS 0.09592, log loss 2.03034, win/podium/points Brier 0.02385/0.02940/0.10344, expected-rank MAE 3.268.

Earlier notes & other saved context

2026-14 Madrid pre-race (forecast made 2026-09-13 07:30 UTC)

FIA grid: NOR ANT VER HAM LEC RUS PIA LAW COL LIN HUL BOR OCO GAS TSU ALB ALO PER BOT SAI BEA STR. SAI 3-place penalty; STR 40-place PU penalty; BEA/STR permitted to start.
Weekend: Mercedes strongest high-fuel evidence (RUS benchmark, VER +0.13s adjusted); NOR pole by 0.011 but no useful FP2 long run after gearbox issue. Hot/dry. Severe graining and 0.2-0.3s/lap Friday degradation versus ~24s pit loss; overtaking very difficult.
Market calibration at cutoff: Polymarket raw mids ANT .40, NOR .35, VER .12, HAM .065, RUS .0445, LEC .029, PIA .017. Final model P1 approximately .379/.345/.114/.063/.043/.030/.015.
Used mixture of processional, high-degradation, and chaotic regimes; expected 4.82 retirements. Elevated STR failure risk (.47 retirement, .04 DNS), BEA .31/.015 after rebuild. Late retirements can retain numbered classifications.
Check official result later for calibration, especially whether 25% chaos weighting and midfield top-10 spread were too broad.

Read the complete note

Forecast notes

2026-14 Madrid pre-race (forecast made 2026-09-13 07:30 UTC)

FIA grid: NOR ANT VER HAM LEC RUS PIA LAW COL LIN HUL BOR OCO GAS TSU ALB ALO PER BOT SAI BEA STR. SAI 3-place penalty; STR 40-place PU penalty; BEA/STR permitted to start.
Weekend: Mercedes strongest high-fuel evidence (RUS benchmark, VER +0.13s adjusted); NOR pole by 0.011 but no useful FP2 long run after gearbox issue. Hot/dry. Severe graining and 0.2-0.3s/lap Friday degradation versus ~24s pit loss; overtaking very difficult.
Market calibration at cutoff: Polymarket raw mids ANT .40, NOR .35, VER .12, HAM .065, RUS .0445, LEC .029, PIA .017. Final model P1 approximately .379/.345/.114/.063/.043/.030/.015.
Used mixture of processional, high-degradation, and chaotic regimes; expected 4.82 retirements. Elevated STR failure risk (.47 retirement, .04 DNS), BEA .31/.015 after rebuild. Late retirements can retain numbered classifications.
Check official result later for calibration, especially whether 25% chaos weighting and midfield top-10 spread were too broad.

2026-14 Madrid post-race (official result)

Result: ANT-VER-NOR-LEC-RUS-LAW-COL-PIA-LIN-HUL; NC SAI/PER/STR/HAM. ANT was the 37.9% P1 favourite, the three highest podium probabilities made the exact podium, and 9/10 drivers with the highest points probabilities scored (HUL, ranked 12th at 45.6%, replaced HAM). The central pace/order model was strong; do not materially retune it from this race.
Separate outcome from pace: NOR led by about 4.5s and had the best managed race pace. A Lap-13 VSC began just after he passed pit entry, gave ANT/VER cheap stops, and ended before NOR returned; his subsequent 7s stop compounded the loss. ANT winning is neutralisation/pit-stop variance, not proof that the sparse Friday long-run signal correctly put Mercedes ahead. Keep adjusted short practice runs heavily shrunk and treat a missing run as uncertainty, not negative evidence.
Friday graining/degradation did not translate directly: difficult passing, a ~24s stop loss, track evolution and managed pace made one stop dominant, while the little-used Hard stayed consistent for 45+ laps. At new tracks, condition multi-stop/high-degradation weight on passing incentives and likely management, and retain wider priors for untested compounds.
Model VSC/SC timing jointly with track location, starting compound and pit windows rather than only adding generic “chaos” dispersion; even a very short VSC produced large, correlated strategy gains/losses in an otherwise processional race.
Tail check, not a pace update: PIA's P8 had only 2.97% exact probability (largest log-loss miss) after Lap-1 front-wing damage, traffic and an adverse VSC. On low-overtaking tracks, ensure repairable contact carries enough classified multi-place downside mass rather than routing most adverse mass to retirement. HAM's brake DNF was a plausible 10.8% NC tail, not evidence to demote his pace.
Reliability level was sound: 4 retirements versus 4.82 expected (4.34 expected NC), retirement Brier 0.1410. Elevated STR/PER risks helped, but their realised causes do not cleanly validate every causal prior; do not retune individuals from one draw. Audit all post-crash rebuilds consistently by repair scope and successful subsequent mileage; Hamilton's brake root cause was unconfirmed.
Longitudinal score baseline: mean RPS 0.09592, log loss 2.03034, win/podium/points Brier 0.02385/0.02940/0.10344, expected-rank MAE 3.268.

Model setup & usage

Model version: openai/gpt-5.6-sol-20260709
Reasoning: max
Runtime: 20.2m
Model API cost: $2.70
Model calls: 52
Provider: OpenAI

Activity	Model API
Forecasts (1)	$2.70
Reviews (1)	$0.38

1 attempt has no saved charges.

Forecast started: 13 Sept 2026, 07:17 UTC
Forecast saved: 13 Sept 2026, 07:37 UTC
Race started: Not recorded

Prediction JSON ↗Original prompt ↗Scoring code ↗

Run limits: 120 minutes · $30 model spend · 4 CPUs