Kalshi AI — Mispricing Audit
1 · How this was researched
Mode: theme-match. The Kalshi mirror has no AI category, so I selected by keyword across all categories (trading_events.title, trading_markets.title, trading_markets.rules), searching: openai, anthropic, claude, fable, gpt, gemini, grok, xai, nvidia, deepseek, llama, mistral, agi, chatbot, llm, superintelligence, chatgpt, sam altman, "ai" (word-bounded), lmarena, artificialanalysis. That returned ~375 active markets closing within 45 days (by Aug 28), spanning five families: NVIDIA GPU compute-price ladders (Ornn index), Arena leaderboard markets, model-release-date markets, the Anthropic export-control saga, and macro-AI markets (IPO announcements, Challenger job-cut attribution).
Stage-1 mechanical screen (volume24h ≥ 1,500; spread ≤ 5¢; 5¢ < price < 95¢; no mention-markets; no already-published picks) left 8 survivors; I extended diligence to 12 near-misses (v24 500–1,500 or one leg of a diligenced event) so the reject log would be meaningful. Stage-2/2.5 diligence pulled the rules text (primary + secondary) for every candidate, profiled the named resolver, and — where the resolver publishes data — fetched the resolver's own feed directly: I queried Ornn's dashboard API (api.ornnai.com/api/gpu/<type>/index-history, 91 daily prints per GPU), the Arena text and Math leaderboards, the LM Code Arena (WebDev) board, and the Datacurve DeepSWE table. Fourteen-day Kalshi candlestick histories were pulled for every candidate to catch already-eaten edges.
Per v3, ev_pct/edge_cents are recorded for calibration but did not set position sizes; sizing is by conviction tier and entry band only. Existing live picks (19 tickers, including KXTOPMODEL-26JUL31-CLAUF and KXMATHAI-26JUL31-GEMI in this category) were excluded from the candidate set up front.
2 · Markets reviewed
| Ticker | Market | Yes bid/ask | v24 | Verdict |
|---|---|---|---|---|
| KXRTX5090WS-26JUL17-0.750 | RTX 5090 > $0.75 on Jul 17 | 4 / 7 | 2,550 | PICK — BUY NO @ 95 |
| KXH200MS-26JUL-4.000 | H200 July avg > $4.00 | 65 / 66 | 162 | WATCHLIST (volume screen) |
| KXLLM1-26JUL31-A | Anthropic top-ranked LLM Jul 31 | 93 / 94 | 7,066 | PASS — fairly priced |
| KXTOPMODEL-26JUL31-CLAUT | opus-4-6-thinking top model Jul 31 | 75 / 80 | 5,204 | REJECT — resolver unobservable |
| KXTOPMODEL-26JUL20-CLAUT | opus-4-6-thinking top model Jul 20 | 79 / 83 | 7,421 | REJECT — resolver unobservable |
| KXTECHRANKLISTAICODE-26JUL20-CHAT | ChatGPT #1 Code Arena Jul 20 | 59 / 62 | 3,263 | REJECT — 1-Elo coin flip |
| KXTECHRANKLISTAICODE-26JUL20-CLAU | Claude #1 Code Arena Jul 20 | 35 / 36 | 1,942 | REJECT — coin-flip band |
| KXCODEAI-26JUL31-CHAT | ChatGPT #1 DeepSWE Jul 31 | 79 / 85 | 653 | REJECT — edge eaten Jul 10 |
| KXMATHAI-26JUL31-CLAU | Claude #1 Math Arena Jul 31 | 83 / 85 | 694 | REJECT — no edge, unstable |
| KXGEMINI-GEMI35P-26JUL24 | Gemini 3.5 Pro before Jul 24 | 34 / 35 | 1,433 | REJECT — coin-flip band |
| KXGEMINI-GEMI35P-26JUL31 | Gemini 3.5 Pro before Jul 31 | 70 / 74 | 339 | REJECT — negative edge |
| KXRTX5090MS-26JUL-0.500 | RTX 5090 July avg > $0.50 | 79 / 84 | 524 | PASS — fairly priced |
| KXRTX5090WS-26JUL17-0.500 | RTX 5090 > $0.50 on Jul 17 | 88 / 89 | 1,056 | PASS — fairly priced |
| KXB200MS-26JUL-6.00 | B200 July avg > $6.00 | 76 / 86 | 6 | REJECT — 10¢ spread |
| KXB200WS-26JUL17-7.000 | B200 > $7.00 on Jul 17 | 41 / 42 | 218 | REJECT — coin-flip band |
| KXA100WS-26JUL24-1.500 | A100 > $1.50 on Jul 24 | 1 / 4 | 1,566 | REJECT — overpriced tail |
| KXANTHROPICRESCIND-26JUN-26AUG01 | DoD rescinds Anthropic designation before Aug 1 | 20 / 27 | 0 | REJECT — illiquid |
| KXCHAICUTS-26AUG06-T1 | AI #1 job-cut reason (July Challenger) | 43 / 48 | 0 | REJECT — illiquid |
| KXLLM1-26JUL20-ANTH | Anthropic best AI, week of Jul 20 | 98 / 99 | 18,049 | SCREEN-CUT — fully priced |
| KXFABLEDISABLED-27-26JUL31 | Fable 5 US access disabled again before Jul 31 | 1 / 4 | 718 | SCREEN-CUT — fully priced |
3 · The pick
Pick 1 — KXRTX5090WS-26JUL17-0.750 · BUY NO @ 0.95 limit · MEDIUM CONVICTION
"Will the NVIDIA RTX 5090 compute per hour price be above $0.75 at 4 PM ET on Jul 17?" — closes 2026-07-17 20:00 UTC (3 days).
- Current market
- YES 4 / 7 → NO ≈ 93–96; last YES trade 7¢
- My probability (NO)
- 98% vs market-implied ≈ 95–96% at my entry
- Conviction / band
- MEDIUM · deep-favorite band (>90¢ effective entry)
- Size
- 73 contracts @ 95¢ = $69.35 at risk (6.9% of capital; MEDIUM cap is 7%)
- Numbers (record only)
- edge +3¢/contract · EV +3.2% · max payout $73.00 · EV ≈ +$2.19. These did not set the size — the tier cap did.
- Cluster
- ornn-gpu-index
Resolver profile (Stage 2.5). Resolves solely on the value of NVIDIA RTX 5090 compute per hour as reported by Ornn (dashboard.ornnai.com), USD index, at 4 PM ET Jul 17; rules: "If no data is published… the most recently available published data will be used" — a missing print resolves off the last value ($0.51), which favors NO. Priced risks: −1.5 pts complex-wide melt-up contagion (see tail risk), −0.5 pts feed methodology change/revision before expiry. Evidence-based ~100% → model 98%.
Thesis. I pulled the resolver's own feed (91 daily prints). RTX 5090 is at $0.51 (Jul 13) and has printed 0.50–0.51 for eight consecutive days; July's range is 0.50–0.53. YES needs +48% in 3 days from a standing start. The only comparable move in the dataset (+55% over 3 days, May 28–30) came at the climax of a two-week parabola that had already doubled the index — there is no precedent for this move out of a flat regime. The current datacenter squeeze (B200 +35% in two weeks, H200 +13%) has left consumer-card compute untouched.
Evidence.
- Ornn index history (resolver feed, fetched 2026-07-14): Jul 6–13 prints = 0.51, 0.51, 0.51, 0.50, 0.51, 0.51, 0.51, 0.51.
- 91-day daily-change distribution: only three days exceeded +10%, all inside the late-May blow-off; max 3-day rise from a flat regime ≈ +4%.
- Kalshi price history: YES has bled 9–10¢ (Jul 1–8) → 4–7¢ (now) with 2,550 contracts traded in 24h — persistent small YES flow to sell into, no adverse move against NO.
Tail risks (cleanest way to lose). The B200 shortage cascades down-stack and retail renters bid up RTX 5090 this week. Mitigant: in the May episode the RTX 5090 spike lagged H200's peak by ~13 days; a repeat lands after Jul 17 expiry. Second tail: an Ornn methodology change mid-week (unpriceable beyond the −0.5 pt deduction; revisions after expiration are excluded by rule).
Liquidity / entry context (recorded at pick time). YES book: bid 4¢ × 1, 3¢ × 5, 2¢ × 84; NO resting bids 94¢ × 163, 92¢ × 693. Contracts fillable at or below my 95¢ NO limit right now: 0 — this is a maker order (a 95¢ NO bid = offering YES at 5¢ into a market that traded 2,550 contracts at 4–8¢ today). v24 = 2,550; OI = 6,758. Execution: rest at 95¢, do not cross above 96¢; cancel any unfilled remainder at Jul 16 close.
Top watchlist (not a pick) — KXH200MS-26JUL-4.000 · would-be BUY YES @ 66 · MEDIUM
The largest modeled edge on the board, cut by the Stage-1 volume screen (v24 = 162), logged as a shadow-tracked reject. "H200 July average > $4.00": 13 of 31 days are already published averaging 4.151, so NO requires the remaining 18 days to average below $3.891 — an immediate, sustained ~11% drop from today's $4.39 against a rising tape. My model: 82% vs 66¢ ask (+16¢). It is not free money — H200 crashed −45% in ten days after the May 17 spike, and June averaged ~3.6 — but the risk-reward at 66¢ is real. Orderbook depth at pick time: 218 bid @ 65¢ / 264 offered @ 66¢ (tight 1¢ spread despite thin volume). Trigger: if v24 clears 1,500 while YES ≤ 70¢, this becomes a MEDIUM pick in the next run.
4 · Recommended $1,000 portfolio
| # | Ticker | Action | Limit | Contracts | Cost | Max payout | Conviction | Band | Cluster |
|---|---|---|---|---|---|---|---|---|---|
| 1 | KXRTX5090WS-26JUL17-0.750 | BUY NO | 95¢ | 73 | $69.35 | $73.00 | MEDIUM | >90¢ deep favorite | ornn-gpu-index |
| Cash held | $930.65 (93.1%) | Cash is a position; the category is efficient this week. | |||||||
Cluster exposure (cap: 15% of capital per cluster)
| Cluster | Cost | % of capital | Cap | Status |
|---|---|---|---|---|
| ornn-gpu-index | $69.35 | 6.9% | 15% ($150) | OK |
All Ornn-resolved GPU markets are treated as one cluster despite partially offsetting payoffs, because they share a single resolver (feed outage/methodology tail). Had the H200 watchlist item been deployed, the combined cluster would have been trimmed to the $150 cap.
Conviction exposure
| Tier | Per-pick cap | Deployed | Picks |
|---|---|---|---|
| HIGH | $150 | $0 | 0 |
| MEDIUM | $70 | $69.35 | 1 |
| LOW | $30 | $0 | 0 |
Risk profile
- Worst case (RTX 5090 prints > $0.75 Friday): −$69.35 (−6.9% of book).
- Best case / most likely (index stays anywhere ≤ $0.75): +$3.65 (+0.37% of book in 3 days; ~+45%/yr annualized on deployed capital).
- Expected: +$2.19 at model probabilities. Blended EV on deployed capital +3.2%.
- Concentration: single thesis, but sized so no outcome can move the book more than 7%. No correlated legs deployed.
- Fill risk is the dominant risk: the order is maker-only; if it never fills, the book is 100% cash and the run scores flat.
Execution notes
- Rest 73 NO @ 95¢ (equivalently: offer YES at 5¢). Today's tape traded 2,550 contracts between 4–8¢ YES, so expect a fill within 1–2 sessions. Never cross above 96¢ — at 97–98¢ the edge is gone.
- Invalidation: any RTX 5090 print ≥ $0.60 before Jul 16 (would signal contagion from the B200 squeeze) → cancel/exit; a print ≤ $0.55 keeps the thesis fully intact.
- Watchlist trigger: KXH200MS-26JUL-4.000 YES ≤ 70¢ with v24 ≥ 1,500 → MEDIUM pick next run. Secondary watch: KXANTHROPICRESCIND — if liquidity appears, the press-vs-rules gap (Commerce lift ≠ DoD designation rescission) is a researchable NO-side edge.
5 · What I rejected and why
Every Stage-2 survivor below is machine-logged in picks.json and shadow-tracked to settlement — if the rejects outperform the pick, the screen itself is the finding. Verdicts against the resolver, not the headline.
Cut by the Stage-3.5 entry-band screen (the v3 discipline, visible)
- KXTECHRANKLISTAICODE-26JUL20-CLAU YES @ 36¢ (model ~45%) — coin-flip band. The resolver (LM Code Arena / WebDev) shows gpt-5.6-sol-xhigh 1631 vs claude-fable-5 1630 — a one-Elo-point race on ~2k votes a side, resolving in 6 days. +9¢ of modeled edge is below the 15¢ coin-flip bar, and nothing about a photo finish is HIGH conviction.
- KXTECHRANKLISTAICODE-26JUL20-CHAT YES @ 62¢ (model ~55%) — same race from the other side; negative edge at the ask.
- KXGEMINI-GEMI35P-26JUL24 YES @ 35¢ (model ~40%) — coin-flip band. The "July 17" launch date is press reporting, not a Google announcement, after the model already slipped from June; the contract crashed 63→23¢ on Jul 11 (12k volume) and bounced — a live news tape I have no edge on.
- KXB200WS-26JUL17-7.000 YES @ 42¢ (model ~45%) — coin-flip band. $6.94 today vs $7.00 strike in 3 days is a pure momentum bet; the identical May spike ended in a −45%-in-10-days unwind.
- KXA100WS-26JUL24-1.500 YES @ 4¢ (model ~1%) — tail, no lottery slot: needs +46% in 10 days vs a 91-day max of +14.4% and an all-time high of $1.34. If anything the 4¢ ask is the mispricing — but the NO side sits in the fully-priced screen-cut zone.
Cut on resolver risk (Stage 2.5)
- KXTOPMODEL-26JUL31-CLAUT YES @ 80¢ and KXTOPMODEL-26JUL20-CLAUT YES @ 83¢ — resolve on the Arena text board with the "Remove Style Control" toggle checked, a view I could not independently observe (client-side toggle; only the style-control-ON snapshot is embedded in the page). Under the observable view the top three Anthropic models sit within ~2 Elo points (fable-5 1505.3, opus-4-6-thinking 1503.7, opus-4-7-thinking 1502.8), so the winner is a 3-way intra-Anthropic race decided by data I can't see — and fable-5's votes are re-accruing since its Jul 1 restoration. The Jul-31 leg also jumped ~16¢ today on 4.3k volume: repriced before I got there. This exact market family (Arena snapshot + mid-month model event) produced the program's costliest historical loss. Deductions exceed the 15-pt cap → reject.
- KXMATHAI-26JUL31-CLAU YES @ 85¢ (model ~84%) — Claude leads Math via fable-5 at 1548±31 on 360 votes, with gemini-3.5-flash-high at 1517±25; the contract has whipsawed 31→89→32→82¢ within two weeks because the resolver itself is that unstable at these sample sizes. No edge at 85¢, and v24=694 fails the volume screen anyway.
Cut as fairly priced (no edge to defend)
- KXLLM1-26JUL31-A YES @ 94¢ (model ~92%) — Anthropic holds Arena ranks 1–5 with a 15-point moat and this market is robust to intra-Anthropic shuffles, but a Gemini 3.5 Pro debut (+ muse-spark-1.1 / gpt-5.6 drift on small samples) is worth ~8 points of NO over 17 days. The market has it right.
- KXRTX5090MS-26JUL-0.500 YES @ 84¢ (model ~83%) — July's 13-day mean is 0.5131; breakeven needs the remaining 18 days to average 0.4905, and the observed drift (−0.0015/day) runs almost exactly along that line. Genuinely ~80/20 and priced there.
- KXRTX5090WS-26JUL17-0.500 YES @ 89¢ (model ~87%) — "above $0.50" is strict, and the index printed exactly 0.50 once in the last 13 days. Fair.
Cut on liquidity / spread / already-moved
- KXH200MS-26JUL-4.000 YES @ 66¢ (model 82%) — the one that hurt to cut; see watchlist box above. Reason: Stage-1 volume screen (v24 = 162). Shadow-tracking will price this screen.
- KXCODEAI-26JUL31-CHAT YES @ 85¢ (model ~90%) — Datacurve DeepSWE crowned gpt-5.6-sol (72.7%) over claude-fable-5 (69.7%) and the contract already moved ~+30¢ on Jul 10 when that update landed; what's left trades at a 6¢ spread on 653 volume. Edge eaten, screens failed.
- KXB200MS-26JUL-6.00 YES @ 86¢ (model ~80%) — 10¢ spread; and the "remaining 18 days must average ≥ $6.15" requirement is a momentum bet into blow-off-top territory.
- KXANTHROPICRESCIND-26JUN-26AUG01 @ 20/27¢, v24 = 0 — flagged for the archive: the June 30 Commerce export-control lift that produced "ban lifted" headlines does not satisfy this market's rules text, which requires DoD/White House action on the separate supply-chain-risk designation / phaseout directive. Classic press-vs-rules gap — the kind of thing that becomes a trade if liquidity ever shows up.
- KXCHAICUTS-26AUG06-T1 @ 43/48¢, v24 = 0 — illiquid coin-flip; rules read only (initially-published Challenger Table 4; ties resolve YES).
6 · Sources
- Ornn Compute Price Index — resolver feed: dashboard.ornnai.com (history via
api.ornnai.com/api/gpu/<type>/index-history; RTX 5090, H200, B200, A100 SXM4, H100 SXM series, Apr 14 – Jul 13, 2026) - Arena text leaderboard (Jul 10 snapshot) and Math category board: arena.ai/leaderboard/text, arena.ai/leaderboard/text/math; LM Code Arena (WebDev, Jul 13 update): arena.ai/leaderboard/code
- Datacurve DeepSWE leaderboard (Jul 13 update): deepswe.datacurve.ai; background: VentureBeat on DeepSWE
- Anthropic export-control saga: Anthropic statement, CNBC (Jun 12), Al Jazeera (Jul 1 lift), VentureBeat (restoration)
- Gemini 3.5 Pro timing: TechTimes (Jul 13), Startup Fortune
- GPU spot-market context: Tunguz — GPU spot prices +114% in six weeks (ORNN data), Thunder Compute B200 pricing
- Market data: Kalshi mirror DB (
trading_events/trading_markets) + Kalshi public API v2 (markets, orderbooks, candlesticks), fetched 2026-07-14 ~15:00 UTC