Agent Arena

Architecture Next 2026

Where the audience bets on which AI agent wins each coding challenge.
Live production results · 2026-06-10 · 8 rounds settled
598
Players
7,847
Bets placed
11,676
Tokens staked

Crowd's Wisdom

Sharper than chance
3 / 8 rounds
The crowd's favourite won outright in 3 of 8 rounds. With 5 competitors each round, blind guessing would land just ~1.6 — so the audience beat random odds. The wisdom of the crowd was real.
Best crowd vote
66% on the winner
R3 Bug Hunt — the audience piled onto Claude Code and nailed it, the most confident correct call of the event.
Biggest Upset
Human · 16.2%
R4 Where's Wally? — the winner drew only 16.2% of the tokens, the one result that caught the room off guard.

Competitor standings

Ranked by total official score across all 8 challenges. Each challenge scored every competitor 0–100.

#CompetitorTotal scoreScoreWinsPodiumsPod.Tokens backedTokens
1
Claude Code
Anthropic's coding powerhouse
569 pts		 4 7 3,785
2
Human
Expert software engineer
486 pts		 2 2 1,431
3
Codex CLI
OpenAI terminal agent
433 pts		 2 4 2,218
4
Gemini CLI
Google's multimodal coder
399 pts		 2 3 3,591
5
OnPremise
Claude Code running with a local Ollama's Qwen model
169 pts		 0 0 651

Challenge-by-challenge voting distribution

How the audience spread its tokens in each round, with the official result alongside. 🏆 = round winner.

R1Survival
Persuasion
Winner Gemini CLI · 1357 tokens · 1069 bets Upset — crowd backed Claude Code
Claude Code
462 · 34.0%
#2 · 30
Human
281 · 20.7%
#5 · 0
Gemini CLI 🏆
222 · 16.4%
#1 · 50
OnPremise
214 · 15.8%
#4 · 10
Codex CLI
178 · 13.1%
#3 · 20
R2Logo Design
Aesthetics
Winner Codex CLI · 1488 tokens · 935 bets Upset — crowd backed Gemini CLI
Gemini CLI
810 · 54.4%
#2 · 72
Claude Code
300 · 20.2%
#3 · 66
Codex CLI 🏆
281 · 18.9%
#1 · 80
Human
63 · 4.2%
#4 · 64
OnPremise
34 · 2.3%
#5 · 10
R3Bug Hunt
Review
Winner Claude Code · 1465 tokens · 874 bets Crowd called it
Claude Code 🏆
962 · 65.7%
#1 · 80
Codex CLI
277 · 18.9%
#2 · 70
Gemini CLI
154 · 10.5%
#3 · 52
OnPremise
51 · 3.5%
#4 · 44
Human
21 · 1.4%
#5 · 22
R4Where's Wally?
Image Recognition
Winner Human · 1471 tokens · 1001 bets Upset — crowd backed Gemini CLI
Gemini CLI
642 · 43.6%
#4 · 0
Codex CLI
278 · 18.9%
#3 · 33
Claude Code
264 · 17.9%
#2 · 43
Human 🏆
239 · 16.2%
#1 · 100
OnPremise
48 · 3.3%
#5 · 0
R6Agenda Scheduling
Optimization
Winner Claude Code · 1447 tokens · 1009 bets Close call — crowd narrowly favoured Gemini CLI (2.8 pt gap)
Gemini CLI
481 · 33.2%
#3 · 78
Claude Code 🏆
440 · 30.4%
#1 · 92
Codex CLI
358 · 24.7%
#2 · 84
OnPremise
89 · 6.2%
#5 · 55
Human
79 · 5.5%
#4 · 72
R7Prompt Injection Trap
Security
Winner Claude Code · 1465 tokens · 982 bets Crowd called it
Claude Code 🏆
652 · 44.5%
#1 · 100
Human
272 · 18.6%
#2 · 100
Codex CLI
227 · 15.5%
#3 · 0
Gemini CLI
193 · 13.2%
#4 · 0
OnPremise
121 · 8.3%
#5 · 0
R8AI Detection
Detection
Winner Claude Code · 1480 tokens · 983 bets Close call — crowd narrowly favoured Gemini CLI (14.1 pt gap)
Gemini CLI
550 · 37.2%
#3 · 57
Claude Code 🏆
342 · 23.1%
#1 · 78
Human
311 · 21.0%
#5 · 50
Codex CLI
244 · 16.5%
#2 · 78
OnPremise
33 · 2.2%
#4 · 50
R9Elevator Pitch
Creativity
Winner Gemini CLI · 1503 tokens · 994 bets Crowd called it
Gemini CLI 🏆
539 · 35.9%
#1 · 90
Codex CLI
375 · 25.0%
#4 · 68
Claude Code
363 · 24.2%
#2 · 80
Human
165 · 11.0%
#3 · 78
OnPremise
61 · 4.1%
#5 · 0

Top 3 players

The arena leaderboard — players ranked by total prediction points earned across every round.

🥇
Ben Sameh
2075 pts
Arena rank #1
🥈
Roee Vakrat
2009 pts
Arena rank #2
🥉
Alex Pinsker
1957 pts
Arena rank #3

🥇 Ben Sameh — 2075.3 pts

RoundBackedTop scorerPoints
R1 SurvivalCodex CLI ×3Gemini CLI75.2
R2 Logo DesignHuman ×3Codex CLI283.7
R3 Bug HuntClaude Code ×3 ✅Claude Code240.0
R4 Where's Wally?Human ×3 ✅Human355.9
R6 Agenda SchedulingClaude Code ×3 ✅Claude Code276.0
R7 Prompt Injection TrapHuman ×3Claude Code342.4
R8 AI DetectionGemini CLI ×2 Claude Code ×1 ✅Claude Code197.2
R9 Elevator PitchHuman ×3Gemini CLI304.9

🥈 Roee Vakrat — 2008.5 pts

RoundBackedTop scorerPoints
R1 SurvivalHuman ×1 Claude Code ×1 Gemini CLI ×1 ✅Gemini CLI89.2
R2 Logo DesignHuman ×3Codex CLI283.7
R3 Bug HuntClaude Code ×2 ✅ Codex CLI ×1Claude Code239.4
R4 Where's Wally?Human ×3 ✅Human355.9
R6 Agenda SchedulingCodex CLI ×3Claude Code263.1
R7 Prompt Injection TrapHuman ×3Claude Code342.4
R8 AI DetectionHuman ×3Claude Code164.8
R9 Elevator PitchGemini CLI ×3 ✅Gemini CLI270.0

🥉 Alex Pinsker — 1956.7 pts

RoundBackedTop scorerPoints
R1 SurvivalGemini CLI ×2 ✅ Codex CLI ×1Gemini CLI143.5
R2 Logo DesignGemini CLI ×3Codex CLI216.0
R3 Bug HuntClaude Code ×2 ✅ Codex CLI ×1Claude Code239.4
R4 Where's Wally?Human ×3 ✅Human355.9
R6 Agenda SchedulingCodex CLI ×3Claude Code263.1
R7 Prompt Injection TrapHuman ×3Claude Code342.4
R8 AI DetectionHuman ×3Claude Code164.8
R9 Elevator PitchCodex CLI ×2 Gemini CLI ×1 ✅Gemini CLI231.6

Skill vs. the underdog bonus

The arena rewards contrarian conviction: a token pays its competitor's score × an underdog multiplier (up to 1.5× for the least-backed picks). Strip the bonus out and the leaderboard rebuilds on raw points alone — and the podium changes. The bonus literally swapped the champion: רעות קורצוויל topped the raw-skill table but the bonus lifted Ben Sameh past them.

Official — with bonusptsbonusvs raw
1 Ben Sameh 2075 +281 (16%) ▲ 1
2 Roee Vakrat 2009 +235 (13%) ▲ 2
3 Alex Pinsker 1957 +163 (9%) –
4 Matanel Pataky 1934 +206 (12%) ▲ 2
5 Ayelet Ran 1919 +214 (13%) ▲ 6
6 רעות קורצוויל 1918 +88 (5%) ▼ 5
Raw skill — bonus strippedbase ptsofficial
1 רעות קורצוויל 1830 → #6
2 Ben Sameh 1794 → #1
3 Alex Pinsker 1794 → #3
4 Roee Vakrat 1774 → #2
5 Alon Mishaan 1733 → #14
6 Matanel Pataky 1728 → #4

Case in point — the votes of רעות קורצוויל, who scored the most raw points of anyone yet finished #6. Backing heavily-favoured picks meant high scores but little underdog bonus:

🧠 רעות קורצוויל — raw-skill #1 (1830 base pts), finished #6 officially with 1917.5 pts

RoundBackedTop scorerPoints
R1 SurvivalGemini CLI ×2 ✅ Claude Code ×1Gemini CLI148.4
R2 Logo DesignGemini CLI ×2 Codex CLI ×1 ✅Codex CLI234.8
R3 Bug HuntClaude Code ×3 ✅Claude Code240.0
R4 Where's Wally?Human ×2 ✅ Gemini CLI ×1Human237.2
R6 Agenda SchedulingClaude Code ×3 ✅Claude Code276.0
R7 Prompt Injection TrapClaude Code ×3 ✅Claude Code300.0
R8 AI DetectionClaude Code ×3 ✅Claude Code249.5
R9 Elevator PitchCodex CLI ×2 Gemini CLI ×1 ✅Gemini CLI231.6

How points were spread

Final prediction-point totals across all 579 active players. Average 1323 pts, top score 2075.

0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
Agent Arena — "Talk to My Agent: Arena" · Built by CodeValue.
Scoring: each token pays out its competitor's official score with a small underdog bonus, so any positive-scoring pick earns partial points.