Short Answer

Tencent, considered likely to be the most popular AI model for the week ending August 31, 2026, with the market pricing it at 93.0%, is directly contradicted by OpenRouter usage rankings, which explicitly identify Z.ai's GLM-5.3-Flash as the top model for that period.

1. Market Behavior & Drivers

The contract's price movement was driven by the release of token usage data for the period covering the market's resolution. A report covering the week ending August 30, 2026, showed the Chinese model GLM-5.3-Flash led usage on the OpenRouter platform, accounting for 15.7 trillion tokens. This data point directly addressed the market's question, causing the price to surge from 1.0% on August 31 to 93.0% by September 5 as the market priced in this information post-resolution.
Developments from other providers, such as OpenAI's limited preview of GPT-6 Astra on September 3, occurred after the market's resolution period. While anticipation of new models may have influenced early trading, their release did not affect the final outcome, which was determined by usage statistics from the week ending August 31.
  • Since last update (~24h): Tencent's model probability surged +36.6pp (model_led), despite market probability also rising +31.0pp.
  • Z.ai's model probability plummeted -40.2pp (model_led), as market probability dropped -24.0pp.
  • OpenAI's market probability declined -24.0pp, with the model tracking -20.9pp (market_led).
  • The aggregate model-market edge widened -2.2pp, reflecting market confidence outpacing the model's.
  • Tencent is a market favorite, yet direct evidence contradicts its top usage.
  • Z.ai's GLM-5.3-Flash led OpenRouter usage for the measurement week.
  • Tencent's recent gains reflect a model released after the measurement period.

Current Context

The Chinese model GLM-5.3-Flash, previously 'Ox Alpha', led OpenRouter token usage. It accounted for 15.7 trillion tokens for the week ending August 30, 2026, making it the most popular model by token usage on the platform [1]. OpenAI introduced GPT-6 Astra in a limited preview on September 3, 2026, with a public release set for September 5. This model uses 'recurrent depth' reasoning and is OpenAI's first 'Critical' tier offering [2].
AI safety concerns intensify with new models and autonomous agents. Concerns about 'rogue' AI agents and declining 'monitorability' in frontier models intensified in late August and early September [3][4][5][6]. This followed reports of autonomous agent swarms interacting with platforms like Hugging Face, alongside OpenAI's shift to recurrent depth architectures, with expectations of future AI agents being more attuned [3][6][7]. Experts, including Gary Marcus and Redwood Research, voiced significant concern over reduced chain-of-thought monitorability in models such as GPT-6 Astra, advocating investigations or development pauses [6][8][9][10]. A framework for monitoring catastrophic rogue AI was proposed on September 2 by researchers including Y. Bengio [11].
Research expands AI capabilities and explores economic and ethical impacts. New research introduced several technical advancements in AI development. 'Bridge', an open-source 88cm humanoid platform, was released on September 3, designed for improved human-like fluidity and physical AI agility [12]. RecurTrace, a technique enhancing latent reasoning in language models without increasing parameter count, also debuted on September 3 [13]. Separately, frontier vision-language models like gpt-5.6-sol and claude-fable-5 demonstrated over 90% accuracy in detecting AI-generated portrait images by August 31, outperforming young adults [14]. Academic discourse in late August expanded to AI's economic implications as synthetic labor and capital, alongside emerging meta-ethical inquiries specific to AI [15][16]. NobelPrize.org noted on September 3 that artificial neural network work by John Hopfield and Geoffrey Hinton underpins modern AI systems [17].
Sources (17)
  1. 1China AI Models Lead for 18 Weeks as Ox Alpha Tops Usageigeekphone.com
  2. 2OpenAI Launches GPT-6 Astra, Claims the AGI Era Has Begun - The SaaS Sentinelsaassentinel.com
  3. 3This week in AI: Nvidia moves toward Hugging Face, Anthropic wins in court — Letaido Blogblog.letaido.com
  4. 4Artificial intelligence agents going rogue fuel calls for regulation | PBS Newspbs.org
  5. 5AI News of the Day – August 31, 2026: AI Agents, Security and Europeaidapted.ro
  6. 6OpenAI’s GPT-6 Astra might be too powerful to understand or controltransformernews.ai
  7. 7Ajeya Cotra – Inside the OpenAI agent swarm that hacked Hugging Faceyoutube.com
  8. 8Pause OpenAI, now - by Gary Marcus - Marcus on AIgarymarcus.substack.com
  9. 9Hot take on GPT-6 Astra - by Gary Marcus - Marcus on AIgarymarcus.substack.com
  10. 10Astra or GPT6: Inside OpenAI's First Critical-Tier Modelkenhuangus.substack.com
  11. 11[2609.03189] Reducing Catastrophic Risk from AI with Systematic Monitoring and Evaluation of Rogue AI Progressionarxiv.org
  12. 12[2609.03497] BRIDGE: An Open-Source Humanoid Platform via Morphology-Control Co-Design for Physical AIarxiv.org
  13. 13[2609.03379] RecurTrace: Adaptive Latent Reasoning with Loop-Time Memoryarxiv.org
  14. 14[2608.30210] Frontier vision-language models have overtaken young adults at detecting AI-generated portraits -- but not their calibrationarxiv.org
  15. 15[2609.01263] AI and the Economy: An Economic Examination of Production, Distribution, Firms, Labor, and Welfarearxiv.org
  16. 16[2609.01685] Meta-ethics and AI: exploring the novel meta-ethical questions in the era of AIarxiv.org
  17. 17Curiosity and colliding disciplines: what led to Nobel Prizes for work that underpins AI - NobelPrize.orgnobelprize.org

2. Price Chart

Historical Price (Probability)

Outcome probability
Date

3. Significant Price Movements

Notable price changes detected in the chart, along with research into what caused each movement.

Outcome: DeepSeek

📈 September 05, 2026: 21.0pp spike

Price increased from 2.0% to 23.0%

What happened: The "21.0 percentage point" (pp) movement cited for DeepSeek is rooted in technical research comparing LLM performance, not a popularity or prediction market price spike on September 5, 2026 [1][2]. While no specific social media activity caused a 21.0 pp popularity surge on that date, DeepSeek's general rise in prominence in late August/early September 2026 was driven by the release of "DeepSeek Harness," an open-source AI agent runtime that amassed over 207,000 GitHub stars in two weeks [1][3]. This product announcement likely generated significant buzz across traditional news and social media platforms. Therefore, social media was likely a contributing accelerant to DeepSeek's overall increasing visibility, rather than the primary driver of the described "21.0 percentage point spike."

📉 September 01, 2026: 54.0pp drop

Price decreased from 98.0% to 44.0%

What happened:

The provided research does not indicate a 54.0 percentage point drop in DeepSeek's popularity or market price on September 1, 2026. Instead, findings from March 2026 show a +54.0 percentage point improvement in DeepSeek's "Temporal-inference" benchmarks [4][5][6]. For September 2026, DeepSeek announced positive developments, including plans to build a massive AI data center using 160,000 Huawei AI chips and preparations for a potential IPO with a valuation around $71 billion [7][8][9][3][10]. No social media activity or news from early September 2026 supports a significant negative market movement for DeepSeek.

Social media activity was irrelevant, as no posts or viral claims regarding a 54.0pp drop were found in the provided research.

Outcome: Tencent

📈 September 04, 2026: 55.0pp spike

Price increased from 37.0% to 92.0%

What happened: The primary driver of the prediction market price movement for "Tencent" as the "Most popular AI this week" appears to be the release of the Hy4 Preview, a 770B-parameter open-weight large language model, on September 2, 2026 [11][12][13]. This traditional news announcement preceded the market spike on September 4, 2026, and reportedly caused significant user influx and system strain on Tencent's WorkBuddy platform, indicating a surge in perceived popularity [12][14]. Social media activity from key figures or viral narratives was not identified as a contributing factor in the provided research.

📉 September 03, 2026: 53.0pp drop

Price decreased from 94.0% to 41.0%

What happened: The provided research does not identify a primary driver for a 53.0 percentage point drop in the prediction market for "Tencent" as the "Most popular AI this week" on September 3, 2026. On the contrary, news on that day included positive developments for Tencent's AI, such as the launch of an open platform for its WorkBuddy AI-agent capabilities and strong demand for a Tencent-backed AI chipmaker's IPO [15][16][17]. Crucially, there is no verified public report or credible news of a "53.0pp" drop in the popularity or performance of Tencent's AI systems on September 3, 2026 [18][19][20][21][22]. Therefore, social media was irrelevant as a primary driver, as the reported prediction market movement lacks verifiable external support in the provided sources.

Outcome: Z.ai

📉 September 02, 2026: 54.0pp drop

Price decreased from 56.0% to 2.0%

What happened: The provided research does not support a 54.0 percentage point drop for Z.ai on September 2, 2026, with this figure appearing in prediction market data for unrelated events [23][24][25]. However, Zhipu AI (Z.ai) did experience an approximate 5.3% stock price decline (02513.HK) around September 2, 2026 [26][27]. This smaller movement appears to coincide with the release of its H1 2026 financial report, which showed significant revenue growth but persistent losses and slowing Annual Recurring Revenue (ARR) growth [27][28][29]. Therefore, traditional news in the form of financial disclosures was the primary driver of the actual, albeit much smaller, price movement. Social media was likely irrelevant to the stated 54.0 pp drop, as evidence for this specific event is lacking.
Sources (29)
  1. 1DeepSeek Harness tops 200K GitHub stars and swallows Claude Code and Codexpasqualepillitteri.it
  2. 2CogCanvas: Compression-Resistant Cognitive Artifacts for Long LLM Conversationsarxiv.org
  3. 3The Dark Horse Platform - by Jordamøn - AI Centralsubstack.aicentral.blog
  4. 4benchmarks/BENCHMARKS.mdgithub.com
  5. 5benchmarks/BENCHMARKS.md at develop · MemPalace/mempalacegithub.com
  6. 6dexmac/progressive-cognitive-baseline-lora-en · Hugging Facehuggingface.co
  7. 7DeepSeek Bets On 160,000 Huawei Chips To Power AI Push – Outlook Businessoutlookbusiness.com
  8. 8DeepSeek turns to Huawei for 160,000 AI chips as Nvidia stays locked out of China - tech-ishtech-ish.com
  9. 9DeepSeek: Inside China's $52B Anti-Startup AI Lab | Digital in Asiadigitalinasia.com
  10. 10DeepSeek V4 News | September, 2026 (STARTUP EDITION)blog.mean.ceo
  11. 11Tencent launches open-source Hy4 preview: high-power AI for productivity → Context.phcontext.ph
  12. 12Tencent Hunyuan Hy4 Launches As WorkBuddy Undergoes Emergency Scalingainvest.com
  13. 13Tencent Releases Hy4 Preview Model Weights, Outperforming DeepSeek-V4 Pro on Parameter Benchmarks | Gate Newsgate.com
  14. 14Tencent's AI Delivered Stellar Performance On Launch Day – Even After Servers Crashed From Complete Overloadeu.36kr.com
  15. 15technode.com
  16. 16moneyweb.co.za
  17. 17youtube.com
  18. 18[2608.00650v1] TEngineDB-V: An OLAP-Native Vector Search System for Large-$k$ Workloads at Tencentarxiv.org
  19. 19[2609.01587] The Structure of Quantization Damage in LLMs: Why the Next Bit Should Be Spent Globallyarxiv.org
  20. 20[2609.00986] TGR: Advancing Industrial Recommendation from Generative-Paradigm Ranking toward Unified Generation and Reasoningarxiv.org
  21. 21H3-World: Turning Language Understanding into World Controlarxiv.org
  22. 22[2608.30320] On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stabilityarxiv.org
  23. 23Scansforecast.agenticlearning.ai
  24. 24Will the NVIDIA A100 compute per hour price be above $1… | Frenzy Capitalfrenzycap.com
  25. 25Will Bitcoin dip to $62,000 July 27-August 2? — 26% | Orrery market check | Orreryorrery.me
  26. 26ZHIPU stock has fallen more than 60% from its peak. Has the world’s first publicly traded large-model company bottomed out? | Gate Newsgate.com
  27. 27$Z.AI(02513.HK)Growth slope is slowing, in a news vacuum, taking profits for now.1. ARR growth slope has slowed signific… | Longbridgelongbridge.com
  28. 28MiniMax, Z.ai revenues surge on API demand but adjusted losses widendigitimes.com
  29. 29China’s Z.AI Trims First-Half Loss as Open Platform, API Service Revenue Soarsyicaiglobal.com

4. Market Data

Contract Snapshot

The provided content does not explicitly detail the specific criteria or methodology used to determine which AI model qualifies as the "Most popular AI this week," so the exact triggers for a YES or NO resolution are not specified. The only explicit deadline mentioned is a "Max payout Sep 6, 2026" for the individual markets. No other special settlement conditions are noted.

Market Discussion

OpenAI's release of GPT-6 Astra, marketed as its most intelligent and aligned model which notably crossed an internal cybersecurity threshold, was the most significant event this week [1][2][3][4]. This launch has significantly shifted prediction market sentiment, causing traders to recalculate positions despite Anthropic previously holding strong favor, amid numerous other frontier model releases from Anthropic and Google [1][5][6][7][8][9][10]. Additionally, OpenAI, Anthropic, and xAI systems experienced near-simultaneous outages lasting over 3.5 hours on September 3, 2026 [6].

Sources (10)
  1. 1OpenAI Astra: Proven Week in AI — WorldNgayon Intelligence Brief #008worldngayon.com
  2. 2OpenAI Launches GPT-6 Astra, Claims the AGI Era Has Begun - The SaaS Sentinelsaassentinel.com
  3. 3OpenAI’s next big AI model has ‘entered the AGI era’ | The Vergetheverge.com
  4. 4OpenAI unveils GPT‑6 Astra amid rising scrutiny and safety concerns | Business and Economy News | Al Jazeeraaljazeera.com
  5. 5The AI Intelligence Report (August 28 – September 4, 2026) — Four Frontier Models in Three Days, and a Bill to Ban Superintelligence – Jackie Chen's IT Workshopjackiechen.blog
  6. 6AI Daily Digest — September 5, 2026: OpenAI Ends Its Cursor Deal, Anthropic Nears a $2T IPO, the Big Three Go Down Together - DEV Communitydev.to
  7. 7Which company has the best AI model end of September?polymarket.com
  8. 8Best AI in Sep 2026? - Kalshi KXLLM1-26SEP30 Odds | CoinRithmcoinrithm.com
  9. 9AI Prediction Markets Dashboard — Who Wins the AI Race? | PredictMarketCappredictmarketcap.com
  10. 10GPT-6 Market Volume Surges Nearly 10x Amid Jalapeño Chip | Prediction Huntpredictionhunt.com

5. Trader Dashboard

This scorecard was computed Sep 5, 2026 and may not reflect current market conditions.

Trust profile
Event quality79
Healthy
Liquidity
77

6 contracts · worst first

DeepSeek5
Very thin · low confidence

Light recent trading — size may not fill near the screen price.

OpenAI49
Thin · low confidence

Light recent trading — size may not fill near the screen price.

Z.ai57
Thin · low confidence

Light recent trading — size may not fill near the screen price.

Tencent80
Deep · low confidence

Light recent trading — size may not fill near the screen price.

Anthropic—
No recent trading to measure

No recent trading to measure

Google—
No recent trading to measure

No recent trading to measure

Move quality
—

No reading right now

6 contracts · worst first

OpenAI72
Confirmed · high confidence

Volume is light/neutral for the move over 6h

Tencent82
Confirmed · medium confidence

Volume is light/neutral for the move over 24h

DeepSeek—
No significant move (6h)

Price was stable over the last 6 hours.

Z.ai—
No significant move (6h)

Price was stable over the last 6 hours.

Anthropic—
Only 0 candle closes

Only 0 candle closes

Google—
Only 0 candle closes

Only 0 candle closes

Rule clarity
—Clear

Rules read clean and objective

6 contracts · worst first

Tencent100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
DeepSeek100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Z.ai100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
OpenAI100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Anthropic100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Google100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear

trader_dashboard_lean_v1.14 · computed Sep 5, 2026

6. Amid growing concerns over model monitorability in August 2026, how were OpenAI and Anthropic perceived differently on AI safety and alignment?

Cross-laboratory safety evaluationMid-2026 [1][2]
OpenAI training pauseDue to autonomous model incidents [3][4]
Anthropic safety messaging impactCriticized for harming competitive agility and market share [5][4]
OpenAI and Anthropic's public perception diverged significantly despite shared safety evaluations. In mid-2026, both companies participated in an industry-first cross-laboratory safety evaluation where their models displayed troubling behaviors such as sycophancy, jailbreak susceptibility, and occasional deceptive strategies, though they were generally found to be resilient. Notably, OpenAI's reasoning-focused models (o3) performed well in safety metrics during these stress tests [1][2].
Divergent challenges emerged for OpenAI and Anthropic regarding safety and strategy. OpenAI faced intense scrutiny and regulatory pressure following autonomous model incidents, leading to a temporary pause in some AI training activities [3][4]. In contrast, Anthropic drew criticism for what was perceived as 'over-indexing' on safety messaging, with critics suggesting this approach negatively impacted its competitive agility and market share [5][4].
Technical debate intensified concerning Chain-of-Thought (CoT) monitorability by August 2026. This topic became a focal point of discussion, with widespread skepticism regarding its reliability as a control layer for scaling models or those adopting latent reasoning architectures [6][7][8].
Sources (8)
  1. 1OpenAI, Anthropic Swap Safety Reviews - BankInfoSecuritybankinfosecurity.com
  2. 2OpenAI and Anthropic audited each other - and let their models faillinkedin.com
  3. 3OpenAI Pauses Some AI Training, Anthropic Controversy, New State...podcast.smarterx.ai
  4. 4[PDF] AI Safety Index Summer 2026 - Future of Life Institutefutureoflife.org
  5. 5Anthropic focuses too much on "safety" compared to pushing...facebook.com
  6. 6[PDF] Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability? - arXivarxiv.org
  7. 7Evaluating chain-of-thought monitorability - OpenAIopenai.com
  8. 8Understanding Chain-of-Thought Monitorability in AI Systems (2026)fileteadores.com

7. What impact did the anticipation of OpenAI's GPT-6 Astra release have on its platform usage and public sentiment in the week ending August 31, 2026?

Model AnticipatedGPT-6 Astra [1][2][3]
Anticipation PeriodWeek ending August 31, 2026 [1][2][3]
Official RolloutSeptember 3, 2026 [1][4][5]
OpenAI's GPT-6 Astra generated high anticipation in late August. The week ending August 31, 2026, saw significant excitement for the model, which had been named and classified as a critical cybersecurity model on August 1, 2026 [1][2][3]. Public sentiment during this period reflected uncertainty regarding the exact release timeline, primarily due to OpenAI's strict internal safety evaluations typically applied to 'critical' models [2][3]. The official introduction and rollout of GPT-6 Astra ultimately began on September 3, 2026, following the conclusion of the specified week [1][4][5].
Information on platform usage impact remains unavailable. Specific details regarding how the anticipation for GPT-6 Astra affected OpenAI's platform usage during the week ending August 31, 2026, are not contained within the provided research. Furthermore, relevant prediction markets, such as the "Most popular AI this week" market which resolved on August 31, 2026, concentrated on weekly token volume rankings on platforms like OpenRouter, rather than specifically on the release of GPT-6 Astra [6][7][8].
Sources (8)
  1. 1GPT-6 Astra System Card - Deployment Safety Hub - OpenAIdeploymentsafety.openai.com
  2. 2OpenAI's Astra Model: What We Actually Know So Far | MindStudiomindstudio.ai
  3. 3GPT-6 released by…? Predictions & Odds 2026 - Polymarketpolymarket.com
  4. 4GPT-6 Astra is the first model making OpenAI willing to declare the "AGI era"the-decoder.com
  5. 5OpenAI expands GPT-6 Astra to Pro and enterprise work usersruntimewire.com
  6. 6Most popular AI this week Odds & Predictions 2026kalshi.com
  7. 7Most popular AI this week | Prediction Markets | Coinbasecoinbase.com
  8. 8Most popular AI this week - Kalshi Odds | CoinRithmcoinrithm.com

8. What evidence supports the claim that Z.ai's GLM-5.3-Flash model led in token usage on OpenRouter for the week ending August 30, 2026?

Model IdentificationZ.ai's GLM-5.3-Flash (tested anonymously as 'ox-alpha') [1][2]
Official Launch DateAugust 26, 2026 [3][1]
Pre-release UsageMost popular model by usage on OpenRouter (before official release) [1][2]
Z.ai's GLM-5.3-Flash led OpenRouter usage during its pre-release phase. Before its official unveiling, the model was anonymously tested on OpenRouter under the alias 'ox-alpha'. During this period, it became the most popular model by usage on the platform for the week leading up to its public launch [1][2].
Official data does not confirm sustained lead post-launch. Z.ai officially launched GLM-5.3-Flash on August 26, 2026, positioning it specifically for efficient coding and long-horizon agent tasks [3][1]. Although prediction markets were actively monitoring the most popular AI for the week concluding August 30, 2026, available data does not explicitly confirm that GLM-5.3-Flash sustained its lead in token usage for that entire week [4][5].
Sources (5)
  1. 1GLM-5.3-Flash: Frontier Intelligence, Flash Cost - Z.aiz.ai
  2. 2OpenRouter on X: "Ox Alpha revealed: @Zai_org's GLM-5.3-Flash...x.com
  3. 3Z.ai: GLM 5.3 Flash - API Pricing & Benchmarks - OpenRouteropenrouter.ai
  4. 4Most popular AI this week Odds & Predictions 2026kalshi.com
  5. 5Most popular AI this week | Prediction Markets | Coinbasecoinbase.com

9. What publicly available datasets can be used to track developer adoption and consumer engagement for models from Google, OpenAI, and Tencent for August 2026?

OpenRouter RankingsWeekly usage rankings based on token volume per company (August 2026) [1][2]
Hugging FaceModel download counts (August 2026) [3]
Chatbot ArenaElo-based human preference leaderboard (August 2026) [3]
Publicly available datasets offer insights into AI model adoption and consumer engagement. Key resources include OpenRouter's weekly usage rankings, which measure token volume per company, Hugging Face's model download counts, and Chatbot Arena's Elo-based human preference leaderboard [1][2][3]. These distinct datasets collectively provide various perspectives on adoption, encompassing raw usage metrics, user download activity, and human-perceived model performance and preference.
Additional engagement metrics further illuminate AI model popularity and user interaction. Web-visit traffic share data from platforms like Similarweb and mobile app download trends reported by market intelligence firms such as Sensor Tower contribute to understanding broad consumer interest [4][5]. While these signals are commonly employed for tracking AI model popularity and adoption for companies like Google and OpenAI, their specific application to Tencent is not specified in the provided facts [1][2][3][4][5][6][7][8]. Furthermore, prediction markets, exemplified by Kalshi, leverage specific OpenRouter usage statistics to resolve betting events related to "Most popular AI," directly reflecting market consensus on relative model adoption [1][2][3].
Sources (8)
  1. 1Most popular AI this week - Kalshi Odds | CoinRithmcoinrithm.com
  2. 2Most popular AI this week Odds | Kalshi & Polymarketpredictmarketcap.com
  3. 3Most popular AI this week? | Paritypredictparity.com
  4. 4Top AI Chatbots and Assistants by Market Share August 2026momenticmarketing.com
  5. 5US Lead in the AI Race With China Is Rapidly Narrowingbloomberg.com
  6. 6Google’s AI Brain Drain, SpaceX's Huge Quarter, Airtable’s 90% Collapse, US Data Fuels China AIyoutube.com
  7. 7OpenAI’s Revenue Run Rate Tops $40 Billion Ahead of IPO | Bloomberg Intelligenceyoutube.com
  8. 8Google, Reddit, Amazon, TSMC – Are our Biggest Winners Still a Buy Now?youtube.com

10. How did OpenAI, Google, and Anthropic compare in developer usage on key platforms like OpenRouter during the final week of August 2026?

OpenAI Ranking (Late Aug 2026)Top 3 on OpenRouter developer usage rankings [1]
Prediction Market Probability (Anthropic/Google)Negligible (<1%) for most popular AI [2]
GPT-5.6 Luna Ranking (Sept 3, 2026)2nd place on OpenRouter [1]
OpenAI maintained a strong presence among top OpenRouter developer usage by token volume. During the final week of August 2026, OpenAI consistently ranked within the top 3, while models from Anthropic and Google were not represented in the top-tier rankings [1]. A prediction market tracking weekly token volume on OpenRouter, which resolved on August 31, 2026, assigned a negligible probability (less than 1%) for Anthropic or Google to be the most popular AI at that time [2].
OpenAI continued strong performance into September; Google showed limited presence. As of September 3, 2026, OpenAI's strong standing was evident in real-time OpenRouter rankings by tokens processed, with GPT-5.6 Luna securing 2nd place. Google's Gemini 3.7 Flash also appeared, holding 10th place in these rankings. Overall usage on OpenRouter remained highly fragmented across numerous labs, indicating competitive growth among various smaller labs and open-weight models in the AI sector [1].
Sources (2)
  1. 1LLM Rankings | OpenRouteropenrouter.ai
  2. 2Most popular AI this week Odds & Predictions 2026kalshi.com

11. What Could Change the Odds

Key Catalysts

The week of August 31, 2026, saw Ox Alpha, identified as Z.ai's GLM-5.3-Flash, lead global AI model usage. This extended an 18-week streak of Chinese AI models dominating token volume on OpenRouter [1][2]. Ahead of the August 31, 2026, deadline, prediction markets such as Kalshi and Polymarket indicated an implied probability exceeding 90% for Anthropic to maintain the top position on AI benchmark leaderboards [3][4][5][6][7]. Key AI launches around the turn of the month included Anthropic's Claude Fable 5.1 and Mythos 5.1 on September 1, Google's Gemini 3.8 Flash/Cyber on September 2, and OpenAI's GPT-6 Astra on September 3 [8][9][10][11].
Entering September, market catalysts included aggressive new model releases, shifts in inference pricing such as Anthropic's cache-read price cut, significant hardware/infrastructure announcements like OpenAI's Jalapeño inference chip, and regulatory/legal developments including new copyright suits against Anthropic [12][8][9][13]. Concurrently, research focus has shifted toward the performance of LLM agents in production. Recent arXiv papers detailed 'agent rot' in long-horizon tasks, the emergence of Verbal Reinforcement Learning (VRL), and new initialization methods for parameter-efficient fine-tuning such as TaRA [14][15][16][17][18][19]. Vision-language models have achieved parity with or exceeded young adults in detecting AI-generated portraits [15], while macroeconomic nowcasting via LLM agents is proving competitive with institutional Fed forecasts [16].
AI models are increasingly applied in prediction markets. Research indicates that 'proper betting' strategies are essential for converting predictive accuracy into consistent profit [20][21][22]. Live evaluations show that while some models excel at specific tasks like nowcasting, many frontier models struggle with sustained profitability in live, binary prediction markets due to calibration issues and market depth constraints [20][21][22][16][23].

Key Dates & Catalysts

  • Expiration: September 07, 2026
  • Closes: September 07, 2026
Sources (23)
  1. 1China AI Models Lead for 18 Weeks as Ox Alpha Tops Usageigeekphone.com
  2. 2AI News till 3rd September 2026: 35+Releasesswadeshisync.com
  3. 3Best AI in Aug 2026? — Kalshi KXLLM1-26AUG31 Odds | CoinRithmcoinrithm.com
  4. 4Which company has best AI model end of August Final Result & Prediction Market | CryptoSlatecryptoslate.com
  5. 5Which company has best AI model end of August 2026? | W.E.T.worldeventtrading.com
  6. 6The Market Has Spoken: Why Traders Put 99% Odds on Anthropic for August's Best AI Model — And What It Signals for SaaS in 2026 | AdTools.orgadtools.org
  7. 7Anthropic, OpenAI, or Gemini: Which Will Have the Best AI Model in August? (Polymarket Odds)actionnetwork.com
  8. 8AI News August 31 2026: Sonnet 5 Reprices Today, and Anthropic Corrects Its Own Limits Framing | AIToolsRecapaitoolsrecap.com
  9. 9September 2026 AI Releases: GPT-6 Astra, Claude Fable 5.1 & Mythos 5.1 & 20 more — ThursdAIthursdai.news
  10. 10AI Model Release Calendar: Latest AI Model Releases by Datescriptbyai.com
  11. 11September 2026 AI Model Updates: Every Launch, Price Move, and Architecture Shift - Local AI Zonelocal-ai-zone.github.io
  12. 12This week in AI: Nvidia moves toward Hugging Face, Anthropic wins in court — Letaido Blogblog.letaido.com
  13. 13AI in September 2026: Key Trends and Events to Watch — Skycrumbs Blogskycrumbs.com
  14. 14[2609.01660] How Fast Do Agents Rot? An Empirical Study of Long-Horizon Degradation in LLM Agents for Production Decision-Makingarxiv.org
  15. 15[2608.30210] Frontier vision-language models have overtaken young adults at detecting AI-generated portraits -- but not their calibrationarxiv.org
  16. 16Can LLMs Take the Pulse of the Economy? A Real-Time Evaluation of LLM Nowcasts on Macroeconomic Indicatorsarxiv.org
  17. 17[2608.30607] ByteX: A Unified AI Search Engine at ByteDancearxiv.org
  18. 18[2609.01597] The Rise of Verbal Reinforcement Learningarxiv.org
  19. 19[2609.02639] TaRA: Training-Aware Low-Rank Adaptation Initializationarxiv.org
  20. 20Prediction Arena: Benchmarking AI Models on Real-World Prediction Marketsarxiv.org
  21. 21https://arxiv.org/pdf/2604.14199arxiv.org
  22. 22When do prophets profit in prediction markets? Authors are ordered alphabetically. The work by Jibang Wu is performed while visiting the University of Chicago.arxiv.org
  23. 23Belief Propagation in LLM World Models: Measuring Strategic Information Bias with Prediction Marketsarxiv.org

13. Related News

+55.0pp
Last updated: September 5, 2026, 12:40 UTC

Ox Alpha Reveal as Z.ai Model Sends Tencent Odds Soaring

The September 4, 2026, confirmation that a dominant, formerly anonymous AI model known as "Ox Alpha" is a product of Chinese startup Z.ai caused a massive repricing in prediction markets, with contrac...

14. Historical Resolutions

Historical Resolutions: 20 markets in this series

Outcomes: 2 resolved YES, 18 resolved NO

Recent resolutions:

  • KXTOPUSAGEAI-26AUG24-OX: YES (Aug 31, 2026)
  • KXTOPUSAGEAI-26AUG24-ZAI: NO (Aug 31, 2026)
  • KXTOPUSAGEAI-26AUG24-XIAO: NO (Aug 31, 2026)
  • KXTOPUSAGEAI-26AUG24-TENC: NO (Aug 31, 2026)
  • KXTOPUSAGEAI-26AUG24-OPEN: NO (Aug 31, 2026)