Scan: AI wins IMO gold medal in 2026?

1. Planner Fable 5.1 · Fermi decomposition + tool plan

Tools chosen: polymarket_direct polymarket_related kalshi_related article_search claude_news gdelt_news wikipedia

Sub-questions (Fermi decomposition)

What is the current Polymarket YES price and volume trend for 'AI wins IMO gold medal in 2026', and what price did the analogous 2025 market trade at and how did it resolve?
Did Polymarket resolve the 2025 version of this market YES based on DeepMind/OpenAI's IMO 2025 gold-level results, or NO because neither the IMO Grand Challenge nor AIMO formally declared a winner — i.e., how strictly are the two named sources applied?
Is the IMO Grand Challenge (imo-grand-challenge.github.io) still active and has it announced or is it expected to announce any AI satisfying its formal-to-formal Lean criteria in 2026?
What is the status and timeline of the AIMO Progress Prize 3 (Kaggle) and the $5M AIMO grand prize — when do results land in 2026, and has any open/publicly-shared model reached gold-medal-level performance qualifying for the grand prize?
Will IMO 2026 (Shanghai, July 2026) run an official AI participation/grading track, and which labs (Google DeepMind, OpenAI, Harmonic, ByteDance Seed, DeepSeek) plan to enter and be officially graded?
Have open-weight models (DeepSeek, Qwen, Kimi, etc.) demonstrated IMO gold-level performance in late 2025/2026, which would matter for AIMO's open-model requirement?
Are there any announcements from AIMO or IMO Grand Challenge organizers in 2026 recognizing an AI as having won the prize/challenge?

Planner reasoning

The core uncertainty is not whether AI can achieve gold-level IMO performance (DeepMind and OpenAI did so unofficially in 2025) but whether the two named resolution sources — the IMO Grand Challenge (formal Lean-based, strict rules) and the AIMO prize (grand prize requires a publicly-shared model hitting gold level on a private set) — will explicitly demonstrate an AI win during 2026, plus how IMO 2026 (Shanghai, July) handles official AI participation. Precedent from how the analogous 2025 Polymarket market resolved is critical for judging how strictly the resolver interprets the sources. Market price and cross-venue signals anchor the estimate.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.5s	1	## This Market's Polymarket Data AI wins IMO gold medal in 2026? - Current price (probability): 27.50% - 7-day price change: +2.50% - 30-day price change: -1.50% - Total volume: $15,965 (USD notional) - Price range: 10.50% - 88.00% - Data points: 90 days
polymarket_related	OK	3.3s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'IMO gold': 0 markets \| keyword 'International Math Olympiad': 0 markets \| keyword 'AIMO': 0 markets \| keyword 'AI math olympiad': 0 markets \| keyword 'IMO 2026': 0 markets
kalshi_related	OK	3.2s	2	2 related markets / summaries. keyword 'IMO gold': ok \| keyword 'math olympiad AI': no matches \| keyword 'AI gold medal': ok
article_search	OK	1.2s	38	Fetched 38 unique articles across 5 queries (lookback=120d). 'AI gold medal International Mathematical Olympiad 2026': 10 articles \| 'AIMO prize progress prize 3 results gold medal': 10 articles \| 'IMO Grand Challenge Lean formal AI': 10 articles \| 'IMO 2026 Shanghai AI participation official gradin
claude_news	OK	38.6s	17	Here are the key findings: AIMO Prize / $5M Grand Prize status: - The $5M AIMO grand prize will be awarded to the first publicly-shared AI model capable of winning a gold medal in the IMO in a future stage of the competition , meaning it has not yet been awarded. Progress Prize 3 (AIMO3) l
gdelt_news	OK	188.2s	0	GDELT: 0 articles across 3 queries (lookback=90d). 'AIMO prize gold medal AI': error GDELT rate-limited after retries (429) \| 'IMO 2026 artificial intelligence gold': error GDELT rate-limited after retries (429) \| 'AI mathematical olympiad 2026': error GDELT rate-limited after retries (429)
wikipedia	OK	0.1s	4	Fetched 4 Wikipedia entries (2 missing pages).

3. Evidence Brief Sonnet 5 · 6767 chars

# Current state The market resolves YES only if the IMO Grand Challenge (strict formal-to-formal Lean certification) or AIMO (formal $5M grand-prize award) formally declares an AI winner during calendar 2026 — not if press/labs merely claim "gold-level" scores. As of now (research post-Sept 2026), no such formal declaration exists from either named source, despite numerous 2026 media claims of AI models achieving gold or perfect IMO scores. # Timeline of key events - 2025-07 (confirmed): Google DeepMind's Gemini Deep Think receives official IMO coordinator grading of 35/42 (gold-level) at IMO 2025 — first officially IMO-graded AI gold, but not certified by AIMO/IMO Grand Challenge. (CBS News) - 2025-07 (reported/self-reported): OpenAI separately claims gold (35/42) via self-organized grading by former medalists, not cooperating with official IMO graders; Harmonic also claims gold. (LessWrong, Inference by Sequoia) - 2025 (confirmed): Analogous Polymarket "AI wins IMO gold medal in 2025?" market resolves NO, since neither AIMO nor IMO Grand Challenge formally declared a winner despite DeepMind's official grading. (claude_news) - 2025-11-19 (confirmed): AIMO Progress Prize 3 (Kaggle, $2.2M) launches; grand $5M prize remains unawarded. (aimoprize.com) - 2025-12 (reported): DeepSeek-V3.2-Speciale claimed as first open-weight model reaching IMO gold-level scores (USCC report) — a benchmark claim, not formal certification. - 2026-07-10 to 07-21 (confirmed): IMO 2026 held in Shanghai; OpenAI has a sponsorship role (ChatGPT Pro for medalists), not a formal AI grading track. (imo-official.org) - 2026-07 (reported/rumored): Multiple gold/perfect-score AI claims emerge — GPT-5.6 Pro (self-reported perfect 6/6), RedNote's dots-note-3.0 (claimed first flawless score, SCMP), others. A mid-2026 tech review counts "two officially graded perfect scores, four self-administered claims, one Lean-formalized proof, one open-weight model at gold level" — underscoring most claims are unofficial. (KuCoin, SCMP, Digital Applied) - 2026 (confirmed, ongoing): IMO Grand Challenge site still frames formal-to-formal Lean protocol as unresolved/in-development; no certified winner announced. (imo-grand-challenge.github.io) # Event Will an AI be formally certified as winning an IMO gold medal (per IMO Grand Challenge or AIMO) between Jan 1–Dec 31, 2026? # Outcomes to forecast Yes / No # Kalshi market anchor No kalshi_direct price was returned in research (gap). Kalshi-related search found no matching IMO/AIMO markets (only unrelated CEO-succession markets). Must rely on Polymarket as closest cross-market proxy: current YES 27.5%, +2.5% over 7d, -1.5% over 30d, volume $15,965 (thin), historical range 10.5%–88% — implying a large repricing down from a high (~88%) likely around/after IMO 2026 concluded without formal certification. # Sub-question answers 1. Polymarket 2026 market: 27.5% current, thin volume (~$16k), 90-day range 10.5–88%. The analogous 2025 market resolved NO despite DeepMind's official IMO gold grading, because AIMO/Grand Challenge didn't formally declare a winner. (polymarket_direct, claude_news) 2. 2025 market resolved NO — sources are applied strictly; official IMO-graded gold (DeepMind) was insufficient without AIMO/Grand Challenge certification. (claude_news) 3. IMO Grand Challenge remains active, still developing a stricter Lean-based formal-to-formal protocol; no AI has met this bar as of 2026. (imo-grand-challenge.github.io) 4. AIMO Progress Prize 3 launched 2025-11-19 on Kaggle; $5M grand prize unawarded; no confirmed timeline for 2026 grand-prize resolution found. (aimoprize.com) 5. IMO 2026 Shanghai (July 10–21) had no official AI grading track; OpenAI's role was sponsorship (ChatGPT Pro giveaway for human medalists), not formal AI entry/certification. (imo-official.org) 6. DeepSeek-V3.2-Speciale (Dec 2025) claimed first open-weight gold-level performance (USCC report); RedNote's dots-note-3.0 claimed a flawless 2026 score — both are unofficial/self-reported, relevant to AIMO's open-model requirement but not certifying. 7. No 2026 announcement found from AIMO or IMO Grand Challenge organizers recognizing any AI as prize/challenge winner. (research silent beyond above) # Key facts (high-confidence, factual) 1. [CBS News] DeepMind's Gemini Deep Think received official IMO 2025 gold-level grading (35/42) — the strongest real-world precedent, but external to AIMO/Grand Challenge. 2. [aimoprize.com] AIMO $5M grand prize unawarded as of research date; Progress Prize 3 ongoing. 3. [imo-grand-challenge.github.io] Formal Lean-based certification protocol still unresolved/in development. 4. [imo-official.org] IMO 2026 Shanghai had no formal AI grading track. 5. [SCMP, KuCoin, Digital Applied] Numerous 2026 gold/perfect-score AI claims exist but are mostly self-reported, not certified. # Cross-market signals - Kalshi related: no matching markets found (irrelevant CEO markets only) — no Kalshi cross-signal available. - Polymarket: 27.5% YES, down sharply from an 88% historical high, thin volume — suggests market recognizes strict resolution criteria are unlikely to be met despite abundant "gold-level" media claims. - Sportsbook implied: none available. # Analyst opinions and speculation - Tech commentary (Digital Applied) frames 2026 as "benchmark saturation" — AI gold-level claims are now common but institutionally uncertified, diluting their evidentiary value for strict resolution sources. - USCC report frames open-model gold-level achievement (DeepSeek) as strategically significant for China's open-AI push, independent of AIMO certification status. # Directional lean per outcome - **Yes**: Multiple 2026 claims of gold/perfect AI performance (OpenAI, RedNote, DeepSeek) suggest technical capability is present; if any lab pursues formal AIMO/Grand Challenge certification, threshold could be met before Dec 31, 2026. - **No**: Both named resolution sources remain unmet as of the timeline of evidence gathered; 2025 precedent shows even official IMO grading wasn't sufficient; IMO Grand Challenge's stricter Lean bar is unmet; AIMO grand prize unawarded; Polymarket pricing collapsed from 88% to 27.5%, reflecting doubt. # Gaps / unknowns - No direct Kalshi YES price captured (tool gap) — must infer from Polymarket proxy. - Unclear whether AIMO or Grand Challenge organizers plan/expect a 2026 certification decision at all. - Uncertain whether any lab is actively pursuing formal Lean-based certification. # Calibration anchors - Polymarket current YES 27.5% (proxy anchor; peaked 88%, troughed 10.5%). - 2025 analogous market resolved NO despite officially-graded AI IMO gold — strong precedent for strict/narrow resolution outcomes even amid positive headlines.

4. Forecasters Fable 5.1 + GPT-5.6 Sol · in parallel

claude-fable-5-1

45% conf

Yes	20%
No	80%

No Kalshi price was available, so the closest anchor is Polymarket at 27.5% YES (thin volume, collapsed from an 88% peak after IMO 2026 concluded without formal certification). Resolution hinges on a formal declaration by the IMO Grand Challenge or AIMO, and neither has certified any AI as of the post-Sept 2026 research date: the Grand Challenge's Lean formal-to-formal protocol remains unresolved and the AIMO $5M grand prize is unawarded. The 2025 analog resolved NO despite DeepMind's officially IMO-graded gold, showing the criteria are applied strictly, and abundant 2026 self-reported 'gold/perfect' claims (OpenAI, RedNote, DeepSeek) do not meet that bar. With only Q4 2026 remaining and no evidence any lab is pursuing formal certification, I shade below the market proxy to ~20% YES, leaving room for a surprise AIMO/Grand Challenge announcement or lenient resolution.

gpt-5.6-sol

72% conf

Yes	16%
No	84%

The closest market anchor is Polymarket at 27.5% YES, but its thin volume and lack of a direct Kalshi price make that signal relatively weak. The 2025 market resolved NO despite an officially graded gold-level performance, demonstrating that unofficial or ordinary IMO grading does not satisfy the narrow certification requirement. Neither AIMO nor the IMO Grand Challenge has announced a 2026 winner, the AIMO grand prize remains unawarded, and the Grand Challenge's Lean protocol remains in development. Multiple gold or perfect-score claims show sufficient technical capability, but without evidence of an imminent formal decision, they do not offset the institutional barriers and limited time remaining.

5. Devil's Advocate Sonnet 5 · red-team critique

Both forecasters shade below the Polymarket 27.5% anchor, but the stated rationale (strict criteria, 2025 NO precedent) is exactly what already caused Polymarket's price to collapse from 88% to 27.5% — using the same evidence to justify a further discount below the proxy risks double-counting the "strictness" factor rather than treating 27.5% as already reflecting it.

Neither forecast explicitly quantifies the narrow remaining time window (research is "post-Sept 2026," leaving only ~3 months before year-end) as a distinct driver; this is a strong mechanical reason for a low probability that deserves separate weighting rather than being folded into vague "no evidence of pursuit" language.

The brief notes "one Lean-formalized proof" among 2026 claims — the closest analog to the Grand Challenge's actual formal-to-formal bar — yet neither forecaster engages with this specific claim or assesses how close it might be to satisfying the strict certification criterion.

Neither considers the tail scenario where AIMO, facing PR pressure from widespread 2026 "gold-level" claims (OpenAI, RedNote, DeepSeek), makes a discretionary/loose certification decision before year-end to stay relevant, which could push YES higher than pure institutional-process reasoning implies.

The confidence values (0.45 vs 0.72) are not well justified given both forecasters used nearly identical evidence and reached similar point estimates (0.20 vs 0.16); this divergence in stated confidence without differentiated reasoning is inconsistent and undermines calibration.

Both treat "No Kalshi price" as a gap but don't flag that Polymarket's thin volume (~$16k) makes the 27.5% anchor itself low-confidence and potentially stale/manipulable, warranting wider uncertainty bands rather than tight convergence near 16-20%.

6. Aggregator Fable 5.1 · final calibrated forecast

19% (58% confidence)

Yes: 19% No: 81%

Resolution requires a formal declaration by the IMO Grand Challenge or AIMO, and as of the post-September 2026 research window neither has certified any AI: the Grand Challenge's Lean formal-to-formal protocol is still unresolved and the AIMO $5M grand prize is unawarded, with only roughly one quarter left in the year. The 2025 analog resolved NO even after DeepMind's officially IMO-graded gold, so the many 2026 self-reported gold/perfect-score claims do not move the needle on the strict criterion. The critique is right that Polymarket's collapse from 88% to 27.5% already prices in much of this strictness, so I shade below the proxy only modestly, mainly for the short remaining window and the absence of any evidence a lab is pursuing formal certification. I retain ~19% for tail paths: a lenient resolver reading, a discretionary AIMO award under PR pressure, or the reported Lean-formalized proof being recognized by the Grand Challenge before year-end.

Pipeline Timing

Total pipeline time: 309.3s

Per-tool research timings shown in the Research section above.