Scan: Will Moonshot be the third-best AI Agent Lab at the end of September 2026?

date	predicted	market price	confidence
2026-09-01	64%	70%	43%
2026-08-23	53%	61%	45%

1. Planner Fable 5.1 · Fermi decomposition + tool plan

Tools chosen: polymarket_direct polymarket_related kalshi_related claude_news claude_news article_search gdelt_news wikipedia

Sub-questions (Fermi decomposition)

What is the current Lab ranking on the arena.ai Agent Arena leaderboard, and where does Moonshot currently sit (excluding AutoEval-marked models)?
How frequently has the third-place lab on the Agent Arena Labs table changed over the past 1-3 months, and which labs have cycled through the top 4?
Has Moonshot announced or is it expected to release a new Kimi agentic model (e.g., Kimi K2/K3 variants) before September 30, 2026 that could move its rank?
Which competing labs (OpenAI, Anthropic, Google, xAI, DeepSeek, Zhipu, Alibaba/Qwen, Meta) have imminent agentic model releases likely to land on the Agent Arena leaderboard before the check date?
How large is the score gap between Moonshot and the labs immediately above and below it on the Agent Arena Labs table, and how much do scores shift with new evaluations?
What are the implied probabilities in the sibling Polymarket markets for third-best lab, and do they sum coherently with Moonshot's price?

Planner reasoning

This resolves on the arena.ai Agent Arena 'Labs' ranking on Sept 30, 2026, so the key questions are the current lab ordering, how volatile the top-3 has been, and whether Moonshot (Kimi) has upcoming agentic model releases or competitors (OpenAI, Anthropic, Google, xAI, Zhipu, DeepSeek, Alibaba) likely to displace it. The Polymarket price is the primary anchor, with sibling markets in the same event group (other labs for 3rd place, plus 1st/2nd-best markets) providing a consistency check.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.4s	1	## This Market's Polymarket Data Will Moonshot be the third-best AI Agent Lab at the end of September 2026? - Current price (probability): 77.55% - 7-day price change: +3.30% - 30-day price change: +44.90% - Total volume: $29,030 (USD notional) - Price range: 23.15% - 82.25% - Data points: 35 d
polymarket_related	OK	3.7s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'third-best AI Agent Lab': 0 markets \| keyword 'Agent Arena': 0 markets \| keyword 'best AI Agent Lab September 2026': 0 markets \| keyword 'Moonshot': 0 markets
kalshi_related	OK	3.5s	1	1 related markets / summaries. keyword 'Agent Arena': no matches \| keyword 'Moonshot Kimi': no matches \| keyword 'AI lab leaderboard': ok
claude_news	OK	34.2s	11	Based on available search results, here are key findings on Moonshot AI's standing on arena.ai's Agent Arena leaderboard: - Lab-level ranking (as of a recent snapshot): Top labs in Agent Arena: #1 OpenAI (GPT-5.5 High), #2 Anthropic (Claude-Opus-4.7 Thinking), #3 Zhipu (GLM-5.1), #4 Google Dee
claude_news	OK	32.9s	11	Based on available search results, here are key findings on Moonshot AI's standing as an "AI Agent Lab" relative to competitors on the Arena.ai Agent Arena leaderboard as of September 2026: - Current Arena Agent Pareto leaderboard (Sep 1, 2026) model ranking: Claude Opus 5 (High) Anthropic lea
article_search	OK	46.8s	35	Fetched 35 unique articles across 4 queries (lookback=45d). 'Moonshot AI Kimi agent model release': 10 articles \| 'Agent Arena leaderboard arena.ai labs ranking': 10 articles \| 'Kimi K2 agentic benchmark ranking': 10 articles \| 'LMArena agent leaderboard top labs': 10 articles
gdelt_news	OK	204.2s	0	GDELT: 0 articles across 3 queries (lookback=30d). 'Moonshot AI Kimi': error GDELT rate-limited after retries (429) \| 'Agent Arena leaderboard': error GDELT rate-limited after retries (429) \| 'Kimi agentic model': error GDELT rate-limited after retries (429)
wikipedia	OK	0.2s	3	Fetched 3 Wikipedia entries (0 missing pages).

3. Evidence Brief Sonnet 5 · 7560 chars

# Current state As of the most recent Agent Arena snapshots (early-to-mid September 2026), Moonshot's Kimi K3 model consistently ranks #4 on the overall Agent Arena leaderboard, effectively #3 by unique lab once Anthropic's two entries (Opus/Fable) are deduplicated. No confirmed, dated "Labs" (rankBy=labs) table snapshot from September 2026 was retrieved directly; the clearest #3-lab evidence (OpenAI>Anthropic>Zhipu>Google>Moonshot #5) predates Kimi K3's July 16, 2026 launch. Resolution depends on the specific "Labs" filtered table on Sept 30, 2026, which may differ from model-level Pareto rankings. # Timeline of key events - 2026-07-16 (confirmed, Wikipedia): Moonshot releases Kimi K3 (2.8T param, largest open-weights model ever). - ~2026-07-20s (reported, x.com/arena): Kimi K3 debuts at #4 on Agent Arena model leaderboard, tied with Claude Opus 4.8 and GPT-5.6 Sol; up from Kimi K2.7 Code's #23. - Pre-K3 snapshot (undated, reported, x.com/arena): Labs order was #1 OpenAI, #2 Anthropic, #3 Zhipu (GLM-5.1), #4 Google DeepMind, #5 Moonshot. - Early Aug 2026 (reported, CryptoSlate): Sibling "best AI Agent end of August" market shows Anthropic 89.5%, OpenAI 9.1%, Moonshot 0.8% — for #1, not #3. - 2026-08-31 (reported, arena.ai/pareto): Claude models lead (+10.58%), GPT-5.6 Sol OpenAI (+9.76%), Kimi K3 Moonshot (+8.74%) — Moonshot #3 by unique lab in Pareto view. - 2026-09-01 (reported, arena.ai/overall/pareto): Ordering: Claude Opus 5 > Claude Fable 5 > GPT-5.6 Sol (OpenAI) > Kimi K3 (Moonshot) > GLM 5.2 (Z.ai) — Moonshot #3 lab (dedup Anthropic). - 2026-09-03 (confirmed, CNBC): Anthropic publicly accuses Moonshot of illegally distilling Claude to build Kimi K3 — reputational/legal risk, not yet affecting leaderboard. - 2026-09-09 (reported, claude_news): Arena "rankBy=labs" page shows 14 labs listed; exact order not retrievable from snippets. - Ongoing (reported): Google Gemini 3.7 Flash added but ranks ~#20 (weak agentic showing); xAI Grok 4.5, Zhipu GLM 5.3, Qwen3.8 Max, DeepSeek v4-pro-high added to Arena but none shown displacing Moonshot in latest snapshots. # Event Will Moonshot rank 3rd among AI Agent Labs on the arena.ai Agent Arena "Labs" leaderboard as checked Sept 30, 2026, 12:00 PM ET? # Outcomes to forecast Yes / No # Kalshi market anchor No direct kalshi_direct data returned. The only same-ticker pricing data comes from the polymarket_direct tool (likely cross-listed/shared market): **current YES price 77.55%**, +3.3% over 7 days, +44.9% over 30 days, volume $29,030, range 23.15%–82.25% over 35 days. Strong upward momentum treat as the consensus anchor. # Sub-question answers 1. **Current Labs ranking / Moonshot's position** — No confirmed live Sept 2026 "Labs" table snapshot retrieved; best model-level evidence (Sept 1, 2026 Pareto) has Anthropic (x2 entries) > OpenAI > Moonshot > Zhipu, implying Moonshot #3 by unique lab (claude_news). An earlier, pre-K3 snapshot had Moonshot #5 with Zhipu #3 (x.com/arena) — now stale. 2. **Frequency of 3rd-place changes** — Evidence suggests volatility: Zhipu held #3 pre-K3; Moonshot displaced it after K3's July 16 launch and subsequent Pareto snapshots (Aug 31, Sep 1) show Moonshot consistently ahead of Zhipu/Google since. No confirmed reversals in Sept. 3. **New Kimi model before Sept 30** — Kimi K3 (July 2026) is the current flagship; no confirmed report of a further K3.x/K4 release before the close date in the research. 4. **Competing labs' imminent releases** — Google's Gemini 3.7 Flash added but underperforms (#20). xAI Grok 4.5, Zhipu GLM 5.2/5.3, Qwen3.8 Max, DeepSeek v4-pro-high all added to Arena but none reported surpassing Moonshot in recent snapshots (claude_news). 5. **Score gaps** — Sept 1 Pareto: Claude Opus 5 +13.74%, Claude Fable 5 +10.61%, GPT-5.6 Sol +9.49%, Kimi K3 +8.71%, GLM 5.2 +6.23%. Gap Moonshot-to-GLM (~2.5pp) is meaningfully larger than Moonshot-to-GPT-5.6 (~0.8pp), suggesting Moonshot's #3(lab)/#4(model) slot is relatively secure vs. #5 but GPT-5.6/OpenAI is close above. 6. **Sibling Polymarket markets** — "Second-best AI Agent Lab end of October": OpenAI 62%, Moonshot 16.5%, Anthropic 13% (implying Moonshot unlikely #2 but plausible #3). "Best AI Agent end of August" (Models filter): Anthropic 89.5%, OpenAI 9.1%, Moonshot 0.8% (for #1). These don't directly price "3rd place" but are consistent with Moonshot being viewed as a strong #3-4 contender, not #1-2. # Key facts (high-confidence, factual) 1. [Wikipedia] Kimi K3 released July 16, 2026, 2.8T params, largest open-weights model ever. 2. [claude_news/arena.ai] Kimi K3 debuted at #4 overall Agent Arena model rank in July, tied with Claude Opus 4.8/GPT-5.6 Sol. 3. [claude_news/arena.ai pareto] Sept 1, 2026 snapshot orders unique labs: Anthropic > OpenAI > Moonshot > Zhipu(Z.ai). 4. [CNBC 2026-09-03] Anthropic publicly accuses Moonshot of Claude distillation — political/legal risk, not yet a leaderboard factor. 5. [claude_news] Google, xAI, DeepSeek, Alibaba/Qwen entries added to Arena but rank well below Moonshot in recent snapshots. # Cross-market signals - Kalshi related: only tangential (labor force participation) markets found; no direct Agent Arena/Moonshot Kalshi comparables. - Polymarket (this market): 77.55% YES, strong +44.9% 30-day uptrend. - Polymarket sibling ("2nd-best lab, end of October"): Moonshot 16.5%, OpenAI 62%, Anthropic 13% — consistent with Moonshot as strong #3, not #1-2. - Polymarket sibling ("best AI Agent end of August," Models filter): Anthropic 89.5% dominant #1, Moonshot only 0.8% — but this is for #1 spot, uninformative for #3. # Analyst opinions and speculation - CNBC/analysts frame Moonshot/Kimi K3 as narrowing the gap with US frontier labs, fueling China AI-race narratives (Hugging Face CEO: China "dominating on open models"). - Distillation allegations (Anthropic vs. Moonshot) could invite regulatory/platform action but no evidence yet of leaderboard delisting. - Token pricing pressure narrative (CNBC 2026-09-01) reinforces Moonshot's commercial traction, indirectly supporting continued high agentic usage/scores on Arena. # Directional lean per outcome - **Yes**: Multiple independent Sept 2026 Pareto/model snapshots place Moonshot's Kimi K3 as #3 by unique lab (behind Anthropic, OpenAI); no rival (Zhipu, Google, xAI, DeepSeek, Qwen) shown surpassing it recently; Polymarket price (77.55%) and steep uptrend align with fundamentals. - **No**: No directly confirmed "Labs"-filtered (vs. Pareto/model) snapshot for late Sept; earlier snapshot had Zhipu at #3; distillation controversy could affect standing/visibility; scores between Moonshot and OpenAI are close (~0.8pp), and Zhipu keeps releasing updates (GLM 5.2/5.3) that could reclaim ground; ambiguity between "Labs" table and "Models" Pareto view per resolution rules adds uncertainty. # Gaps / unknowns - No live, dated "Labs" (rankBy=labs) table snapshot for September 2026 was retrieved — the exact resolution-relevant view. - Whether AutoEval-flagged entries (e.g., DeepSeek) affect ordering unclear. - No confirmed Kimi K3.x/K4 release plans before Sept 30. - True kalshi_direct price for this ticker not returned in tool output; anchor price sourced from polymarket_direct. # Calibration anchors - Current YES price (anchor): 77.55% (polymarket_direct, only pricing data available), +44.9% 30-day trend. - Precedent: Rapid reordering occurred once before (Zhipu #3 → Moonshot #3-4) within ~6 weeks of a major model launch (Kimi K3, July 2026), showing 3rd place is contestable but currently favors Moonshot.

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

claude-fable-5-1

55% conf

Yes	77%
No	23%

The Polymarket anchor sits at 77.55% YES with strong upward momentum, and the fundamentals align: Sept 1 Pareto snapshots order unique labs Anthropic > OpenAI > Moonshot (Kimi K3, +8.71%) > Zhipu (GLM 5.2, +6.23%), giving Moonshot a ~2.5pp cushion over #4 while none of the recently added rivals (Google, xAI, DeepSeek, Qwen) rank near it. Downside risks are that Moonshot could overtake OpenAI (only ~0.8pp gap, which would resolve No), Zhipu's GLM 5.3 could reclaim ground, or the exact 'Labs' table view differs from the Pareto/model view since no dated Labs snapshot was retrieved. With only ~3 weeks until resolution and no confirmed disruptive releases, I stay close to the market consensus.

gpt-5.6-sol

68% conf

Yes	77%
No	23%

The 77.55% YES market price is the best available consensus anchor, and the evidence broadly supports it. August 31 and September 1 snapshots imply Moonshot is third by unique lab, while its roughly 2.5-point lead over GLM is substantial and no newly added competitor is reported to have overtaken it. The lack of another confirmed Kimi release is not especially concerning because K3 already holds the relevant position, whereas competing releases from Google, Zhipu, xAI, Qwen, and DeepSeek have not produced a documented reversal. I remain slightly below the raw evidentiary case's apparent strength because the resolution-specific Labs table was not directly retrieved and may rank labs differently from the model-level Pareto view.

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

Both forecasters flag the critical ambiguity—no confirmed "Labs" (rankBy=labs) table snapshot exists, only Pareto/model-level views—yet both land at the identical 77.55% anchor price without independently stress-testing how a "Labs" aggregation (which could weight by multiple models per lab, e.g., Anthropic's Opus+Fable, or Zhipu's GLM 5.1/5.2/5.3 entries) might systematically differ from the Pareto snapshot used as their main evidence. Neither considers that a labs-level aggregation could plausibly boost Zhipu (multiple GLM entries) or OpenAI relative to Moonshot's single K3 entry, which is a distinct risk from simple score-gap analysis.

Both forecasters cite the "~0.8pp gap to OpenAing" as a key downside risk that could flip Moonshot from #3 to #4, but neither models what happens if Moonshot instead moves UP to #2 (displacing OpenAI) — under the stated resolution criteria ("3rd place"), that scenario also resolves No, yet neither forecast explicitly separates "Moonshot falls to #4/#5" from "Moonshot rises to #2" as distinct paths to No, potentially understating the true No probability given Moonshot's momentum (largest open-weights model, closing gap narrative).

Neither forecast substantively engages with the Anthropic distillation controversy (CNBC 2026-09-03) beyond noting it as "not yet affecting leaderboard" — but if Arena or its partners take punitive action (delisting, disqualification, flagging as derivative work) before Sept 30, this is a discrete tail risk that could remove Moonshot from contention entirely, which is categorically different from a mere ranking slip and deserves more explicit weight.

The sibling Polymarket market ("second-best AI Agent Lab end of October": Moonshot 16.5%, OpenAI 62%, Anthropic 13%) is presented as corroborating evidence, but neither forecaster reconciles why a market pricing Moonshot at only 16.5% for #2 one month later is consistent with 77.55% confidence for #3 now — arguably the sibling market suggests more downside crowding around Moonshot's rank stability than the anchor price implies.

Both forecasts converge suspiciously close to the market price (77.7% vs 77.55%) with minimal independent adjustment despite acknowledging real ambiguity (undated Labs table, distillation risk, close OpenAI gap) — this looks like anchoring bias rather than genuine synthesis, since the stated uncertainties (list of at least 3 distinct downside scenarios) arguably warrant a wider spread or a probability further from the raw market consensus in either direction.

Neither forecaster addresses the +44.9% 30-day momentum as potentially reflecting stale information catch-up (market repricing after K3's July debut) versus forward-looking confidence about the Sept 30 resolution specifically — momentum from a market that also lacks a confirmed "Labs" snapshot could itself be built on the same ambiguous Pareto data, making it weaker corroboration than treated.

6. Aggregator Fable 5.1 · final calibrated forecast

75% (62% confidence)

Yes: 75% No: 25%

The most recent model-level Pareto snapshots (Aug 31, Sep 1) consistently place Moonshot's Kimi K3 as the #3 unique lab behind Anthropic and OpenAI, with a ~2.5pp cushion over Zhipu and no newer entrant (Google, xAI, DeepSeek, Qwen) reported anywhere near it, which aligns with the 77.55% market anchor and its strong uptrend. The critique correctly notes two distinct No paths that both forecasters underweighted: Moonshot rising to #2 past OpenAI (only ~0.8pp gap) also resolves No, and the resolution-specific 'Labs' table was never directly observed and could aggregate differently from the Pareto view; the distillation dispute is a small additional tail. The sibling October market (Moonshot 16.5% for #2) is actually consistent with Moonshot sitting at #3, so it does not argue for a large downward move. I shave slightly below the market to reflect the unverified Labs view and the two-sided nature of the risk, landing at 75% Yes.

Pipeline Timing

Total pipeline time: 302.0s

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