Tyler Jacobsma is the founder of Flowframe.xyz, which provides in-depth content and tools for prediction market traders.
The Polymarket market "Which company has the #1 AI model at the end of April? (Style Control On)"
asks who will top the LMArena Style Control leaderboard at noon Eastern on April 30.
Anthropic is the favorite at 90¢ – Yes, meaning the market is pricing an 10% chance someone else takes the top spot in the next 13 days. I think the real number is 13–17%, a small edge, but an edge is what we look for, nonetheless.
Position: Buy No on Anthropic at 10¢. Fair value: 13–17¢. Edge: +18% to +55% return on a 13-
day hold. Key date: April 30, 12:00 PM ET (not midnight).
Use Polymarket invite code Action to get early access and bypass 1 million+ people on the waitlist. Our Polymarket invite code also unlocks a trading bonus: Deposit $20, Get $20 Bonus!
What "Style Control" Actually Means
LMArena lets users blind-vote on pairs of model answers. Style Control is a setting that strips out formatting tricks (long answers, bullet lists, bold text) before scoring, so a model has to win on substance rather than presentation.
Claude has historically done well here. GPT models have done worse because their default style leans on the formatting that Style Control removes.
Polymarket Odds
The Current Leaderboard
The April 14 snapshot, the most recent refresh as of April 17:
| Rank | Model | Company | Score | CL | Votes |
|---|---|---|---|---|---|
| 1 | Claude-Opus-4-6-Thinking | Anthropic | 1502 | ±5 | 17,219 |
| 2 | Claude-Opus-4-6 | Anthropic | 1496 | ±5 | 18,377 |
| 3 | Muse-Spark (Preliminary) | Meta | 1495 | ±9 | 4,182 |
| 4 | Gemini-3.1-Pro-Preview | 1493 | ±5 | 21,708 | |
| 5 | Grok-4.20-Beta1 | xAI | 1485 | ±6 | 10,884 |
| 6 | GPT-5.4-high | OpenAI | 1481 | ±6 | 10,633 |
Anthropic owns #1 and #2. The closest non-Anthropic model is Meta's muse-spark at 1495, but that score is flagged "Preliminary" with only 4,182 votes, roughly half the sample of every other model on the board. Preliminary scores tend to drift down, not up, as more votes come in.
Claude Opus 4.7 was released on April 16 and has not yet hit the leaderboard. Anthropic's release notes show 4.7 beating 4.6 on 12 of 14 internal benchmarks. The previous 4.5-to-4.6 jump was about 29 Elo points.
If 4.7 lands anywhere near 1502, Anthropic's lead over the rest of the board widens from 7 Elo to 15–25.
Polymarket's parallel "Claude Opus 4.7 Arena Debut ≥1480 score" market is at 95% Yes, which is independent confirmation that the market expects 4.7 to land strongly.
Why the No Side Is Mispriced
The 10¢ price implies a single number: 10% chance someone besides Anthropic finishes #1 on April 30. Break that 10% into the actual threats and it stops looking high enough.
OpenAI "Spud" / GPT-5.5 (~7%). Pretraining finished March 24, per The Information. Greg Brockman has called it "big model feel, not an incremental improvement." OpenAI quietly retired GPT-5.2-Codex and GPT-5.1-Codex-Mini on April 14, the kind of housekeeping that usually precedes a launch. Polymarket's "Spud release by April 30" market trades at 75–80%.
The hard part is Style Control. Current GPT-5.4-high sits at 1481, a 21-point gap to Claude. Spud needs to land and clear 1502 on Style Control specifically and gather enough votes in under a week. Stack the conditional probabilities, and you get 7–9%.
Google (~3–5%). Google I/O is May 19–20, three weeks after the market resolves. Shipping a flagship before I/O sacrifices the marketing moment, and Google rarely does that. Gemini 3.1 Pro Preview has been on the board since February at 1493 and has never overtaken Claude. Possible, not likely.
Meta (~2–4%). Muse-spark at 1495. Preliminary is the closest threat by score, but Preliminary ratings usually settle lower, not higher. The alphabetical tiebreaker also costs Meta any exact tie with Anthropic.
Adding it up with some overlap correction: roughly 13–17%. The historical base rate supports this. Between August 2025 and April 2026, the #1 slot on LMArena changed hands twice across 17 two-week windows, a 12% rate.
Risks
The thesis breaks if my 15% number is too high. The strongest version of that argument: Opus 4.7 hits the leaderboard within days, lands at 1515+, and pushes the bar for any challenger so high that the realistic dethroning probability drops to 7–9%.
At that point, the 10¢ price is fair, and the trade is a wash.
The right hedge is to wait two to four days for 4.7's debut Elo before sizing in. If 4.7 lands above 1515, skip the trade.
If it lands at 1505–1512, the original thesis gets stronger.
Bottom Line
Anthropic is the right favorite as 90¢ is defensible. The 10¢ No side sits slightly under a 12% historical base rate.
OpenAI's potential launch of Spud is a catalyst that nudges the real NO probability up to around 15%, in my opinion.
Take the No Anthropic position after confirming Opus 4.7's debut Elo lands below 1515.
