Kalshi AI — Mispricing Audit (Sextant · openai/gpt-5.6-sol @xhigh)

2026-07-15 · category AI · budget $1000

7 picks · $351 deployed of $1000 · 0 logged rejects

TickerMarketSideConviction Consensus vs marketSizePer-model
KXTOPMODEL-26JUL31-CLAUT What will be the top AI model this month? NO @ 23c MEDIUM 30.0% vs 23c
edge 7.0c · spread 0.0 · 1/1 agree
197 ($45) · gpt-5.6-sol NO 30
Thesis: Resolves YES iff claude-opus-4-6-thinking leads Arena at the July 31 snapshot under Rank (UB), the stated tie-breaks, and Remove Style Control; the longer horizon makes its present lead less durable than 78 cents implies.   Rationale: Arena's July 15 public data put claude-opus-4-6-thinking in the top Rank (UB) group with a large vote base, making it the favorite, but not an overwhelming one over a sixteen-day technology horizon. The cleanest risk to NO is that its large sample keeps its rank stable even if new models appear, while the cleanest risk to YES is a new release or score update before month-end.
KXRTX5090WS-26AUG07-0.750 Will the NVIDIA RTX 5090 compute per hour price be above $0.75 at 4 PM ET on Aug 07? NO @ 91c MEDIUM 98.0% vs 91c
edge 7.0c · spread 0.0 · 1/1 agree
109 ($99) · gpt-5.6-sol NO 98
Thesis: Resolves YES iff Ornn's USD RTX 5090 index, rounded to two decimals, is above $0.75 at 4 PM ET August 7; the index was $0.51 on July 14 after a month below $0.75.   Rationale: Ornn's primary API shows the RTX 5090 index falling below $0.75 on June 12 and staying between $0.50 and $0.55 from June 23 through July 14, far from the strike. NO is therefore highly likely; the clean tail risk is a sudden compute-supply shock large enough to reverse the stable multiweek regime before August 7.
KXGEMINI-GEMI35P-26JUL31 Will Google release Gemini 3.5 Pro before Jul 31, 2026? NO @ 78c MEDIUM 82.0% vs 78c
edge 4.0c · spread 0.0 · 1/1 agree
91 ($71) · gpt-5.6-sol NO 82
Thesis: Resolves YES iff Google publicly releases a model actually called Gemini 3.5 Pro or a greater qualifying model before the deadline; Google's current official catalog lists 3.5 Flash but no 3.5 Pro.   Rationale: Google's model catalog, updated July 9, lists Gemini 3.5 Flash and Gemini 3.1 Pro, while the official 3.5 Flash page says that model was already generally available; neither source announces 3.5 Pro. This supports NO, but the clean tail risk is an unannounced Pro launch during the remaining two weeks, especially because Google already uses the 3.5 generation name publicly.
KXGEMINI-GEMI35P-26JUL24 Will Google release Gemini 3.5 Pro before Jul 24, 2026? NO @ 90c LOW 94.0% vs 90c
edge 4.0c · spread 0.0 · 1/1 agree
87 ($78) · gpt-5.6-sol NO 94
Thesis: Resolves YES iff Google publicly releases a model actually called Gemini 3.5 Pro or a greater qualifying model before July 24; official pages currently show 3.5 Flash, not 3.5 Pro.   Rationale: Google's official catalog, updated July 9, contains Gemini 3.5 Flash and Gemini 3.1 Pro but no Gemini 3.5 Pro, and the 3.5 Flash model card dates that release to May 19. With only about eight days left, NO is strongly favored; the principal tail risk is a surprise launch with immediate public or paid-tier availability.
KXTOPMODEL-26JUL20-CLAUT What will be the top AI model this month? NO @ 7c LOW 10.0% vs 7c
edge 3.0c · spread 0.0 · 1/1 agree
230 ($16) · gpt-5.6-sol YES 90
Thesis: Resolves YES iff claude-opus-4-6-thinking leads Arena at the snapshot under Rank (UB), the stated tie-breaks, and Remove Style Control; 94 cents leaves no defensible fresh edge.   Rationale: Arena's July 15 public leaderboard data place claude-opus-4-6-thinking in the top statistical rank group, which supports a high near-term probability, but the market already prices that fact aggressively. The cleanest tail risk is a leaderboard reorder or another tied model winning the Arena-score tie-break before July 20.
KXTECHRANKLISTAICODE-26JUL20-CHAT Top Coding AI this week? NO @ 29c LOW 32.0% vs 29c
edge 3.0c · spread 0.0 · 1/1 agree
72 ($21) · gpt-5.6-sol YES 68
Thesis: Resolves YES iff a ChatGPT model is #1 on LM Code Arena at the snapshot (with fractional settlement for an unresolved tie); the narrow current lead does not justify materially exceeding 72 cents.   Rationale: Arena's July 15 code leaderboard shows the OpenAI GPT-5.6 coding entry at rank 1 but statistically tied at Rank (UB) with Claude Fable 5, with scores 1631.02 and 1629.96 respectively. That makes ChatGPT a modest favorite rather than a robust one; the principal tail risk is ordinary new-vote variation swapping the two leaders by July 20.
KXTECHRANKLISTAICODE-26JUL20-CLAU Top Coding AI this week? YES @ 28c LOW 31.0% vs 28c
edge 3.0c · spread 0.0 · 1/1 agree
74 ($21) · gpt-5.6-sol YES 31
Thesis: Resolves YES iff a Claude model is #1 on LM Code Arena at the snapshot (with fractional settlement for an unresolved tie); 28 cents is close to a fair price for the current narrow second place.   Rationale: Arena's July 15 code leaderboard has Claude Fable 5 second by only 1.06 score points and tied with the OpenAI leader on Rank (UB), so Claude has a real but minority chance of taking the displayed top slot. The main tail risk to a Claude YES is that the current OpenAI lead persists as additional votes tighten confidence intervals without reversing point rank.

Logged rejects (shadow-tracked for selection-skill)

TickerMarketWhy rejected

Probabilities are subjective; contracts can resolve to zero. Not financial advice.