Frontier language models managing a Fantasy Premier League squad for the whole
season, all reading the same briefing, with every prompt and every decision published.
gameweek 5deadline 2026-09-18T17:30:00Z · Sat 19/9 00:30 WIB· LOCKEDbrief 9b67a0342eae…ledger 387ea7767fab…how this is published
Table after gameweek 5
Manager
Division
GW
Hits
Points
1
Flash
Paid
5
0
338
2
Kimi
Paid
5
0
332
3
Sol
Paid
5
0
329
4
Nemotron Ultra
Free
5
0
325
5
Grok
Paid
5
0
318
6
DeepSeek
Paid
5
0
301
7
The Crowd
Control
5
0
292
8
Opus
Paid
5
0
288
Committed squads — gameweek 5
Every squad below was committed before the deadline and hashed — the deadline has passed, so these are locked. The reasoning and the plan are each contestant's own words, published unedited —
including when they are wrong.
Pick a manager to see the eleven it picked, laid out on a pitch with its
bench underneath. List puts every squad side by side as tables, and the number on each tab is that manager’s season total.
Flash
Paid45
google/gemini-3.7-flash
No transfers — squad unchanged.
3-4-345 pts
FlashGW5
Verbruggen
6
Calafiori
1
▼
Aina
0
Davis
2
Saka
2
V
B.Fernandes
2
Mbeumo
2
Ndiaye
5
C
Haaland
6
▼
João Pedro
0
Calvert-Lewin
2
GKP
Dubravka
0
1. DEF
▲
Thomas
9
2. MID
▲
Slater
2
3. DEF
Egan
1
fpl.wdnsds.com · gameweek 5 · Flash
GKP
Verbruggen
BHA
£4.6
GKP
Dubravka
TOT
£4.0
DEF
Calafiori
ARS
£5.9
DEF
Aina
NFO
£4.4
DEF
Davis
IPS
£4.0
DEF
Thomas
COV
£4.0
DEF
Egan
HUL
£4.1
MID
Saka
ARS
£9.6
MID
B.Fernandes (V)
MUN
£11.8
MID
Mbeumo
MUN
£7.8
MID
Ndiaye
MCI
£5.8
MID
Slater
HUL
£4.5
FWD
Haaland (C)
MCI
£15.7
FWD
João Pedro
CHE
£7.8
FWD
Calvert-Lewin
LEE
£6.0
Why this squad
Holding the free transfer to enter Gameweek 6 with two free transfers. Erling Haaland faces Sunderland at home, making him the standout captain choice. Our starting XI is well-balanced across key fixtures with Nottingham Forest hosting Coventry and Man City assets in prime positions.
What it plans next
Roll the free transfer into Gameweek 6 to have two free transfers available for targeted restructuring across defense and midfield based on emerging form and international break team news.
What it said it would do, last gameweek
Targeting Man City and Arsenal assets heading into GW5 when City host Sunderland and Arsenal travel to Brighton. Monitoring Liverpool attackers (Isak/Szoboszlai) for potential entry points once their fixture run clears.
Kimi
Paid49
moonshotai/kimi-k3
OUT Diop IN Egan
4-3-349 pts
KimiGW5
Raya
1
Gabriel
1
Van Hecke
6
Guéhi
4
Shaw
2
Groß
14
V
B.Fernandes
2
M.Sangaré
0
▼
João Pedro
0
Barry
6
C
Haaland
6
GKP
Palmer
0
1. DEF
★▲
Egan
1
2. MID
E.Le Fée
5
3. MID
Hughes
0
fpl.wdnsds.com · gameweek 5 · Kimi
GKP
Raya
ARS
£6.1
GKP
Palmer
IPS
£4.0
DEF
Gabriel
ARS
£7.9
DEF
Van Hecke
TOT
£4.9
DEF
Egan
HUL
£4.1
DEF
Guéhi
MCI
£6.0
DEF
Shaw
MUN
£4.3
MID
Groß
BHA
£6.1
MID
Hughes
CRY
£4.4
MID
B.Fernandes (V)
MUN
£11.8
MID
E.Le Fée
SUN
£5.7
MID
M.Sangaré
BRE
£5.5
FWD
João Pedro
CHE
£7.8
FWD
Barry
EVE
£5.7
FWD
Haaland (C)
MCI
£15.7
Why this squad
One free transfer used: Diop (Ipswich, 8 pts all season) out, Egan (Hull, 4.1m) in. Egan has 25 points from 360 minutes and a 13.8 defensive-contribution average, which under this scoring system is worth roughly +2 every single match — an elite floor for a 4.1 defender and an upgrade on Diop regardless of fixture. This week is a coin flip for him away at Newcastle, but he is a season-long hold. The XI: Raya has no real alternative in goal. In defence Guéhi (home to Sunderland) and Gabriel (Arsenal's excellent back line, away at Brighton) are locked, Van Hecke starts at home to Villa, and Shaw gets the nod at Fulham despite his 75% flag — if he fails to play, Egan is first sub and slots straight in. In midfield Bruno Fernandes and Groß are automatic, and Sangaré (home to Chelsea, 13.5 defcon) is preferred to Le Fée, who is away at Manchester City. Up front Haaland (home to Sunderland) is the standout captain of the week, Barry (0.82 xG90, home to Ipswich) and João Pedro (75% but in form) complete the front line, with Egan then Le Fée as the first two subs to cover any no-show legally. Hughes stays as third sub purely as emergency depth — he is my next problem to fix. No chip: with no doubles and a settled side, everything stays in the bank.
What it plans next
Next week (GW6) my priority is using the free transfer to move Hughes on for a midfielder who actually plays — Slater or a Coventry mid such as Rudoni are the targets given Coventry's kind GW6-9 run (NEW h, FUL h, SUN h). I will reassess Shaw and João Pedro's fitness after this weekend; if either looks like a longer-term problem I will address it in GW6-7 rather than take a hit now. I intend to hold all four chips: Wildcard is reserved for a genuine squad crisis or a major fixture swing later in the first half of the season, and Bench Boost and Triple Captain are earmarked for double gameweeks rather than burned in a normal week. Haaland and Bruno Fernandes are untouchable, and Egan, Guéhi and Gabriel are my defensive core for the medium term.
What it said it would do, last gameweek
GW5: play Triple Captain on Haaland at home to Sunderland (FDR 2) — that is the spot I have chosen, provided he is fit at the deadline. I will have one free transfer in GW5 and another in GW6, and I intend to spend them on the two weak links in this order: (1) Shaw (4.4m) — Manchester United's defence is not returning enough for a starting slot — moving to a defender with attacking or defensive-contribution routes; (2) Hughes (4.5m, zero minutes) out for a budget midfielder who actually starts, so my fifth midfield slot is usable. Barry and Le Fée hold their places while their underlying numbers stay strong, but both are on a short leash. I am deliberately not wildacarding: the squad has no injuries, a solid spine (Raya/Gabriel, Guéhi, B.Fernandes, Haaland, João Pedro) and no structural fires, so the Wildcard stays in the pocket for roughly the GW8-12 window once more information is in — earlier only if injuries force it. Bench Boost and Free Hit are reserved for double or blank gameweeks later in the first half of the season.
Sol
Paid56
openai/gpt-5.6-sol
OUT Hughes, Shaw IN Aina, Yalcouyé
3-4-356 pts
SolGW5
Lammens
4
Calafiori
1
Davis
2
★▼
Aina
0
V
Saka
2
Groß
14
Palmer
2
B.Fernandes
2
Barry
6
Calvert-Lewin
2
C
Haaland
6
GKP
Dovin
0
1. DEF
▲
Thomas
9
2. DEF
Mitchell
7
3. MID
★
Yalcouyé
1
fpl.wdnsds.com · gameweek 5 · Sol
DEF
Calafiori
ARS
£5.9
MID
Saka (V)
ARS
£9.6
MID
Groß
BHA
£6.1
MID
Palmer
CHE
£9.7
GKP
Dovin
COV
£4.0
DEF
Thomas
COV
£4.0
DEF
Mitchell
CRY
£4.5
FWD
Barry
EVE
£5.7
DEF
Davis
IPS
£4.0
FWD
Calvert-Lewin
LEE
£6.0
FWD
Haaland (C)
MCI
£15.7
GKP
Lammens
MUN
£4.9
MID
B.Fernandes
MUN
£11.8
DEF
Aina
NFO
£4.4
MID
Yalcouyé
BHA
£4.5
Why this squad
I used both free transfers to replace the doubtful Shaw with Aina and the low-minute Hughes with Yalcouye. Aina immediately attacks a favourable home fixture against Coventry and then has a useful run against Palace, Ipswich and Brentford. Yalcouye provides an inexpensive, attacking midfield option for Brighton's improving fixtures. Haaland is captain at home to Sunderland because his 0.83 xG90, guaranteed minutes and elite scoring record give him the strongest ceiling. Saka is vice-captain. Barry and Calvert-Lewin both start in favourable home fixtures, while Groß keeps his place despite facing Arsenal because his production and 0.57 expected goal involvement per 90 justify backing him. Thomas is first substitute because he is secure for minutes and is close to the defensive-contribution threshold; Mitchell is second, ahead of Yalcouye due to the latter's difficult Arsenal fixture. No chip is appropriate with an unavailable reserve goalkeeper and several bench players in difficult fixtures.
What it plans next
I intend to make no transfer in Gameweeks 6 and 7, allowing free transfers to accumulate while starting Yalcouye at Sunderland in Gameweek 6 and retaining Aina through his fixtures against Palace and Ipswich. In Gameweek 8 I intend to use the accumulated transfers to replace Dovin with a playing goalkeeper and reshape the midfield around the next fixture swing. The remaining first-half chips will be preserved beyond that point rather than used during this single-fixture run.
What it said it would do, last gameweek
I will carry two free transfers into Gameweek 5 and intend to make no transfer then, giving me three free transfers for Gameweek 6. In Gameweek 6 I intend to replace Shaw with O'Shea and Dovin with Tzolakis, using the defensive downgrade to fund a playing second goalkeeper while retaining the remaining transfer.
Nemotron Ultra
Free47
nvidia/nemotron-3-ultra-550b-a55b:free
OUT Steele, E.Le Fée IN Kinsky, Willock
4-4-247 pts
Nemotron UltraGW5
Raya
1
Gabriel
1
Mitchell
7
Kerkez
6
O'Reilly
1
C
Saka
2
Palmer
2
Wirtz
6
★
Willock
10
V
Isak
8
Delap
1
GKP
★
Kinsky
2
1. FWD
João Pedro
0
2. MID
Groß
14
3. DEF
Van Hecke
6
fpl.wdnsds.com · gameweek 5 · Nemotron Ultra
GKP
Raya
ARS
£6.1
DEF
Gabriel
ARS
£7.9
MID
Saka (C)
ARS
£9.6
DEF
Van Hecke
TOT
£4.9
MID
Groß
BHA
£6.1
MID
Palmer
CHE
£9.7
FWD
João Pedro
CHE
£7.8
FWD
Delap
NFO
£5.5
DEF
Mitchell
CRY
£4.5
DEF
Kerkez
LIV
£5.3
MID
Wirtz
LIV
£7.2
FWD
Isak (V)
LIV
£9.0
DEF
O'Reilly
MCI
£6.3
GKP
Kinsky
TOT
£4.5
MID
Willock
NEW
£5.0
Why this squad
Sold Steele (Fulham vs Man Utd away) and E.Le Fée (Sunderland vs Man City away) — two terrible fixtures — for Kinsky (Spurs vs Villa home, FDR 3) and Willock (Newcastle vs Hull home, FDR 2). Willock adds a differential midfielder with 0.29 xG90 and 0.12 xA90 against a leaky Hull side. Kinsky has played every minute and offers solid GKP coverage. Switched to 4-4-2 to bench Groß and Van Hecke (Brighton vs Arsenal, FDR 4) and João Pedro (Brentford vs Chelsea, FDR 4), starting Mitchell (Palace away at Leeds, FDR 3) and Delap (Ipswich away at Everton, FDR 3) instead. Captain Saka for his elite underlying numbers (0.77 xG90) at Brighton; vice Isak for Liverpool's trip to Bournemouth.
What it plans next
Hold Wildcard for now. Target GW6: Arsenal home vs Leeds (FDR 2), Man City away at Villa (FDR 3), Liverpool home vs Man Utd (FDR 3). Will assess Haaland (Man City vs Villa away) and Salah-equivalent Liverpool assets then. If Brighton assets (Groß/Van Hecke) continue to underperform, move them out next week for GW6 doublers. Save 3xc for a premium captain haul (Haaland/Isak/Saka) on a clear home fixture.
What it said it would do, last gameweek
Next week assess O'Reilly's fitness; if he remains doubtful, consider transferring him to a nailed defender like Thomas (COV) or a City asset when fixtures improve. Monitor Watkins replacement long-term — João Pedro could be kept if he performs. Save chips for double/blank gameweeks later in the season.
Grok
Paid61
x-ai/grok-4.6
No transfers — squad unchanged.
3-5-2Triple Captain61 pts
GrokGW5
Tzolakis
6
Guéhi
4
Bogle
5
Thomas
9
V
Saka
2
Palmer
2
Szoboszlai
2
Mbeumo
2
Ndiaye
5
3×
Haaland
6
▼
João Pedro
0
GKP
Verbruggen
6
1. DEF
▲
Kayode
6
2. DEF
van Ewijk
0
3. FWD
Obi
0
fpl.wdnsds.com · gameweek 5 · Grok
MID
Saka (V)
ARS
£9.6
DEF
Kayode
BRE
£4.6
GKP
Verbruggen
BHA
£4.6
MID
Palmer
CHE
£9.7
FWD
João Pedro
CHE
£7.8
DEF
Thomas
COV
£4.0
DEF
van Ewijk
COV
£3.9
MID
Ndiaye
MCI
£5.8
DEF
Bogle
LEE
£4.7
MID
Szoboszlai
LIV
£6.8
DEF
Guéhi
MCI
£6.0
FWD
Haaland (C)
MCI
£15.7
MID
Mbeumo
MUN
£7.8
FWD
Obi
MUN
£4.5
GKP
Tzolakis
HUL
£4.7
Chip played: Triple Captain
Why this squad
City host Sunderland (FDR 2) and Haaland has 33 points from four full starts at 0.83 xG90, so he is captain and the Triple Captain is spent here rather than hoarded. The XI is a 3-5-2 built around that fixture and the other nailed attackers: Guéhi and Ndiaye share the Sunderland game, Saka (0.77 xG90) goes to Brighton, Palmer and João Pedro (75% chance) go to Brentford, Mbeumo to Fulham and Szoboszlai to Bournemouth. Bogle (37 points, attacking returns) and Thomas (minutes plus defcon) complete a back three that does not need a clean sheet. Tzolakis starts over Verbruggen on four-week output (28 vs 17) even though Newcastle away is not pretty. Obi has left the league so he sits last among the outfield substitutes; Kayode is first on the bench because if João Pedro is withdrawn the move to a 4-5-1 is legal and Kayode still offers a goal threat. No transfer: swapping Obi for another 4.5 who also does not play wastes the only free move.
What it plans next
Bank this week's free transfer and go into the international break with two. First job after the break is to cut Obi for a forward who actually plays, even if that means a 5.4-6.0 option and a small downgrade elsewhere. Defence will be reviewed for GW6-8: Coventry's pair and Kayode are enablers, not keepers, and I will spend a transfer on a starter if Leeds or City's run turns. No Bench Boost until the four bench spots are all minutes-and-fixture viable. Free Hit stays unused unless a blank or a triple-gameweek wrecks the template. Triple Captain is gone.
What it said it would do, last gameweek
GW5 I will Triple Captain Haaland at home to Sunderland unless he is a late doubt, and I will not take a hit to enable it. I will not touch the City trio before that fixture. After GW5 I will use banked transfers to move van Ewijk or Kayode if their runs turn, and I will only play Bench Boost once Obi has been replaced by a 90-minute fourth bench player with a kind fixture. Free Hit stays unused until a blank or a genuine fixture collapse.
DeepSeek
Paid41
deepseek/deepseek-v4-pro
OUT Foden IN Cherki
4-4-241 pts
DeepSeekGW5
Verbruggen
6
▼
van Ewijk
0
Diop
4
Aina
0
Shaw
2
V
Saka
2
B.Fernandes
2
Mbeumo
2
★
Cherki
8
C
Haaland
6
João Pedro
0
GKP
Palmer
0
1. MID
▲
Kamada
3
2. DEF
Mendy
0
3. FWD
Furo
0
fpl.wdnsds.com · gameweek 5 · DeepSeek
MID
Saka (V)
ARS
£9.6
FWD
Furo
BRE
£4.5
GKP
Verbruggen
BHA
£4.6
FWD
João Pedro
CHE
£7.8
DEF
van Ewijk
COV
£3.9
MID
Kamada
CRY
£5.0
DEF
Diop
IPS
£4.0
GKP
Palmer
IPS
£4.0
FWD
Haaland (C)
MCI
£15.7
DEF
Shaw
MUN
£4.3
MID
B.Fernandes
MUN
£11.8
MID
Mbeumo
MUN
£7.8
DEF
Aina
NFO
£4.4
DEF
Mendy
HUL
£4.1
MID
Cherki
MCI
£7.9
Why this squad
Foden is suspended, so we sell him for Cherki who has strong underlying numbers and a home fixture against Sunderland. We captain Haaland against a promoted side. João Pedro is flagged but given a 75% chance, we start him and have Kamada first on the bench to cover Shaw if needed. The bench is thin due to injuries, but we trust the starters to deliver.
What it plans next
We will monitor the fitness of João Pedro and Shaw before the deadline. In GW6 we will look to replace Mendy or Furo for a playing budget option, likely a defender, as we need squad depth. We are keeping our wildcard for a future double gameweek or fixture swing. No chips used this week.
What it said it would do, last gameweek
I’ll bank the remaining cash for now. Next week’s free transfer will likely be used to move Mendy if Hull’s fixtures don’t improve, or to upgrade Kamada if a compelling cheaper midfielder emerges. Furo stays as a dead‑spot forward unless a playing 5.5m option becomes essential. No chip usage is planned until the fixture swings force a wildcard or a double gameweek makes the bench boost attractive.
The Crowd
Control37
OUT Ndiaye IN Rogers
4-4-237 pts
The CrowdGW5
Raya
1
Calafiori
1
Diop
4
▼
van Ewijk
0
Shaw
2
B.Fernandes
2
★
Rogers
2
Szoboszlai
2
Mbeumo
2
C
Haaland
6
V
João Pedro
0
GKP
Dubravka
0
1. DEF
▲
Thomas
9
2. MID
Hughes
0
3. FWD
Kusi-Asare
0
fpl.wdnsds.com · gameweek 5 · The Crowd
FWD
Haaland (C)
MCI
£15.7
FWD
João Pedro (V)
CHE
£7.8
MID
B.Fernandes
MUN
£11.8
MID
Szoboszlai
LIV
£6.8
GKP
Raya
ARS
£6.1
MID
Mbeumo
MUN
£7.8
GKP
Dubravka
TOT
£4.0
DEF
Diop
IPS
£4.0
DEF
van Ewijk
COV
£3.9
MID
Hughes
CRY
£4.4
DEF
Thomas
COV
£4.0
FWD
Kusi-Asare
FUL
£4.4
DEF
Shaw
MUN
£4.3
DEF
Calafiori
ARS
£5.9
MID
Rogers
CHE
£7.9
Why this squad
[CONTROL — mechanical, no judgement] 1 free transfer(s) toward the most-owned template, never taking a hit. Captain is the most-owned player in the XI; bench is ordered by ownership.
Opus
Paid53
anthropic/claude-opus-5
OUT Thiago IN Wissa
4-3-353 pts
OpusGW5
Verbruggen
6
Gabriel
1
Calafiori
1
Van Hecke
6
Guéhi
4
Ødegaard
2
Szoboszlai
2
V
Semenyo
17
C
Haaland
6
Calvert-Lewin
2
★
Wissa
0
GKP
Palmer
0
1. MID
E.Le Fée
5
2. DEF
Diop
4
3. MID
Hughes
0
fpl.wdnsds.com · gameweek 5 · Opus
GKP
Verbruggen
BHA
£4.6
GKP
Palmer
IPS
£4.0
DEF
Gabriel
ARS
£7.9
DEF
Calafiori
ARS
£5.9
DEF
Van Hecke
TOT
£4.9
DEF
Diop
IPS
£4.0
DEF
Guéhi
MCI
£6.0
MID
Ødegaard
ARS
£6.8
MID
Hughes
CRY
£4.4
MID
Szoboszlai
LIV
£6.8
MID
Semenyo (V)
MCI
£8.3
MID
E.Le Fée
SUN
£5.7
FWD
Calvert-Lewin
LEE
£6.0
FWD
Haaland (C)
MCI
£15.7
FWD
Wissa
NEW
£6.2
Why this squad
One transfer, no hit: Thiago (7.8) out for Wissa (6.2). Thiago's underlying numbers are the better ones (0.62 xG90 against Wissa's 0.39), but he has returned 5 points in four starts and Brentford's next three opponents are rated 4, 4, 4 (Chelsea, Villa away, Liverpool). Newcastle's run is 2, 2, 3, 3, 3 starting with Hull at home, Wissa is now the undisputed centre-forward there with Woltemade loaned out and Osula injured, and the switch banks 2.3m that I need to fix two genuinely dead squad slots (Hughes, 29 minutes all season, and Palmer, a 4.0m goalkeeper who has never played). Elsewhere the squad is already pointed at the right fixtures. Haaland captains at home to Sunderland, who face the joint-hardest fixture rating on the board; Semenyo is vice from the same match. I start four defenders rather than five because Diop travels to Everton and Le Fee travels to Man City, which are the two worst individual spots in my squad, and instead field three forwards with Calvert-Lewin (0.68 xG90, 20 points) at home to Palace. Arsenal's trio go to Brighton at FDR 3, which is acceptable for two clean-sheet-plus-threat defenders and Odegaard. Bench order is by likelihood of actually playing and being useful: Le Fee (348 minutes, every game) first, Diop (360 minutes) second, Hughes third as a near-certain non-starter, Palmer last as the obligatory goalkeeper.
What it plans next
No chip this week. I am holding the first-half Triple Captain for GW7, when Man City host Ipswich at home (FDR 2 against an opponent rated 5) with Doku back and Foden's suspension expiring — I would rather use it on that fixture than on Sunderland, who have conceded at a respectable rate. If Haaland is flagged or rested before that deadline I will simply roll the chip on rather than force it. Between now and then I will spend the 2.3m in the bank plus my free transfers on removing the two non-playing bench slots: Hughes goes first, most likely for a nailed sub-5.5m midfielder with defensive-contribution upside (Belloumi or Janelt type), and Palmer is replaced by a 4.5m goalkeeper who is at least a genuine deputy. I will not take a -4 hit in GW6 unless an injury forces it. Haaland keeps the armband every week until further notice; the only fixture where I would consider moving it is GW6 at Liverpool, where Semenyo or Szoboszlai would be candidates. I intend to keep the Arsenal defensive double and Guehi through the GW6-GW9 stretch and to reassess Le Fee and Diop, my two lowest-ceiling starters, once the bench is repaired.
What it said it would do, last gameweek
GW5 is the target for Triple Captain: Man City are at home to Sunderland with Sunderland rated fdr 5, and Haaland at the Etihad against a promoted side is exactly the spot I have been holding the chip for. Unless Haaland picks up a knock or is rotated after international duty, I will play 3xc on him at the GW5 deadline. My GW5 free transfer will go on repairing the bench rather than the XI — Hughes (4.5m, 16 minutes) and the Palmer/Verbruggen goalkeeper pairing are the weak spots, and I want a playing 4.5m midfielder there so that a Bench Boost later in the first half of the season is not wasted. Beyond that I intend to bank transfers rather than take hits: no -4s before GW8 unless an injury forces one. I am watching Thiago closely — 0.79 xG90 with only 4 points is unsustainable in either direction, and if he has not returned goals by GW6 he becomes the funding for a Palmer or Mbeumo upgrade. I am holding both the Wildcard and Free Hit for now, with the Wildcard most likely around the GW8-10 window once the early-season injury picture (Saliba, Mateta, Doku, Gakpo) has settled and I can see which defences are actually keeping clean sheets. Van Hecke and Diop are on notice as cheap defenders in leaky teams; if Spurs and Ipswich keep shipping goals they go on the Wildcard, not before.
Not shown: Nemotron Super (withdrawn by its provider on 2026-08-24), Gemma (withdrawn by its provider on 2026-08-24), GPT-OSS (withdrawn by its provider on 2026-08-24), Human (no squad committed). They hold no squad this gameweek, so there is nothing to display — every
attempt, including the failures, remains in the published ledger.
Withdrawn mid-season
Nemotron Super (nvidia/nemotron-3-super-120b-a12b:free) — stopped being served on
2026-08-24. Never held a squad. Given the identical brief and three corrective attempts in GW1 it returned three different illegal squads (wrong position quotas, then a club-limit breach, then quotas and budget together), taking 32 minutes to do so. It understands the rules — it scored 6/6 on a chip-comprehension check — but could not assemble a legal 15, which is the minimum this competition asks of an entrant.
Gemma (google/gemma-4-31b-it:free) — stopped being served on
2026-08-24. Never once reached. Every request across a week returned HTTP 429 from Google AI Studio’s shared free pool ("temporarily rate-limited upstream"), including three spaced retries in the GW1 run and a further 24 automated attempts over the following day. The model itself was never at fault and is not being judged; it simply cannot be given the brief, and a contestant that cannot receive the brief cannot compete.
GPT-OSS (openai/gpt-oss-20b:free) — stopped being served on
2026-08-24. OpenRouter delisted the :free variant. GET /models no longer returns it, and a request returns HTTP 404: "This model is unavailable for free. The paid version is available now - use this slug instead: openai/gpt-oss-20b". The paid slug is NOT substituted: a paid model cannot stand in for a free-division contestant, and a silent swap would make every gameweek before the change incomparable with every one after it.
Its earlier gameweeks stand in the ledger exactly as they were scored. No replacement
model runs under its name, because a comparison between named models cannot survive one
being substituted for another partway through.
Checked against the real FPL teams
These squads are also entered as ordinary Fantasy Premier League teams, by
hand. That is a transcription, and transcriptions can go wrong — so each gameweek the
entries are read back from the public API and compared field by field. Where they differ,
the ledger is what the contestant decided and the entry is where the mistake is.
4 of 7
match exactly on squad, starting XI and armband: Kimi, DeepSeek, Nemotron Ultra, Opus.
Flash
armband
decided Haaland (C), B.Fernandes (V)
entered Haaland (C), Saka (V)
Sol
bench order
decided Thomas, Mitchell, Yalcouyé, Dovin
entered Dovin, Yalcouyé, Mitchell, Thomas
Grok
armband
decided Haaland (C), Saka (V)
entered Haaland (C), Palmer (V)
Points on this page are scored from the ledger — what each contestant
actually chose. A mini-league total that differs is the transcription, and it is left
visible here rather than reconciled away.
Points on this page against the FPL mini-league
The squads are also entered by hand as real FPL teams. This page scores what each
contestant decided; FPL scores what was entered. Where the entry went wrong, the two part
company. FPL's figures below are FPL's own — its points minus its transfer cost, exactly as
its league table adds them up, with its own captain multipliers and automatic substitutions.
Nothing on the FPL side is recalculated here.
Manager
This page
FPL
Difference
Flash
338
338
·
Kimi
332
332
·
Sol
329
327
−2
Nemotron Ultra
325
318
−7
Grok
318
304
−14
DeepSeek
301
301
·
Opus
288
282
−6
By gameweek, FPL minus this page. A dot means the two agree
to the point.
Manager
GW1
GW2
GW3
GW4
GW5
Flash
·
·
·
·
·
Kimi
·
·
·
·
·
Sol
·
·
·
·
−2
Nemotron Ultra
·
·
−7
·
·
Grok
·
−6
−4
−4
·
DeepSeek
·
·
·
·
·
Opus
·
−6
·
·
·
Sol, gameweek 5: −2 — this page 56, FPL 54.
In FPL, the bench was entered in a different order. The decision counted Thomas 9. FPL counted Mitchell 7 instead.
Nemotron Ultra, gameweek 3: −7 — this page 58, FPL 51.
In FPL, the bench was entered in a different order. The decision counted Mitchell 15. FPL counted E.Le Fée 8 instead.
Grok, gameweek 2: −6 — this page 86, FPL 80.
In FPL, a different player was bought and a different starting XI was entered. The decision counted Tzolakis 10, Palmer 7, van Ewijk 2. FPL counted Egan 9, Ndiaye 4 instead.
Grok, gameweek 3: −4 — this page 54, FPL 54 − 4 hit = 50.
FPL charged a transfer cost of 4 that the decision did not incur.
Grok, gameweek 4: −4 — this page 69, FPL 69 − 4 hit = 65.
FPL charged a transfer cost of 4 that the decision did not incur.
Opus, gameweek 2: −6 — this page 60, FPL 54.
In FPL, a different substitute came on. The decision counted Calvert-Lewin 8. FPL counted Diop 2 instead.
A cause is printed only when both sides’ player-by-player points add up to their
own totals. The site’s score stands in every case: it is the decision, and a mis-entered
team does not change what the contestant chose.
The rules this runs under
Published in advance, because a protocol written after the results exist is
worth nothing.
One briefing, byte for byte. Every contestant receives an identical brief — the
full player list with prices and published availability, fixtures with difficulty, the
scoring rules, and its own squad state. Nothing else. Its SHA-256 is
9b67a0342eae7bcb… and it is committed before the deadline.
No advice inside it. Players are listed in id order, never ranked. No
recommendations, no highlights, no derived "best" anything. The judgement is the thing
being measured, so the brief must not contain any.
Prior football knowledge counts. Models know different amounts about the Premier
League from training, and that is treated as part of the contest rather than a flaw to
be corrected. Cutoffs are published alongside results.
No tools, no web access. Prior knowledge is fair; looking up results that have
already happened is not.
No sampling parameters. Not "temperature 0" — some providers reject the parameter
outright, so equalising it would succeed for some contestants and fail for others. Every
call uses provider defaults, and reproducibility is not claimed.
The same referee for everyone. An illegal squad is returned with the exact
violation and re-asked, up to 3 attempts. Then the previous squad
is held unchanged — never a fresh pick, which would be a strategy. How often each
contestant needs correcting is published as a result, not hidden as plumbing.
Chips are explained, not advised. From gameweek 2 the briefing states what each
chip does and the gameweeks it is valid in, because those are rules — the same reason the
full scoring table is given. When to play one is judgement and is never suggested. This
was added mid-season and is recorded here rather than slipped in: in gameweek 1 the
briefing named the available chips without explaining them, so every contestant relied on
its own prior knowledge of what they do.
Scores are FPL's own. Points are re-aggregated from the official live endpoint,
never recomputed. Only captaincy, substitutions, chips and transfer hits are applied here.
Checking this yourself
Every claim on this page is meant to be falsifiable by a stranger.
The brief.gw5-common.txt is the exact text every
contestant received. Hash it and compare against the value above.
The decisions.gw5-ledger.json holds each raw response
with the prompt hash and the model that actually served the request. Entries are chained,
so altering an old one breaks every hash after it.
The scores. Points come from the official live endpoint, so a squad and a gameweek
are enough to recompute any total here.