Open vs. Closed Token Volume (open weights share) in Oct 26, 2026
Short Answer
Open-weight models are expected to process above 52.4% of gateway tokens by October 26, 2026. The September 2026 Vercel AI Gateway report already shows a 71.5% share, with the market pricing this at 99.0%.
1. Market Behavior & Drivers
The rapid repricing of this market appears driven by a cascade of high-frequency data from AI model routers showing a dominant token share for open-weight models. The final spike to 99% on October 26 coincides with reports of OpenRouter production traffic from the previous day, where open-weight models accounted for 79.6% of tokens. This single data point provided strong, recent evidence that reinforced earlier reports, such as Vercel AI Gateway's September data showing a 71.5% token share for open models.
The preceding climb from 75% to 85% on October 25 also tracked reporting from multiple independent sources showing open-weight token volume ranging from 60% to nearly 80%. No specific news event or data release has been identified that corresponds with the initial 15.0 percentage point spike on October 23. The market's overall trajectory reflects a swift assimilation of production traffic data that shows open-weight model usage far exceeding the contract's likely resolution threshold.
- Open-weight token share likely exceeds 57.8%, per Vercel's September 2026 report.
- Vercel AI Gateway data reports 71.5% open-weight share for September 2026.
- This 71.5% share indicates the 74.0% threshold is unlikely to be met.
Who Wins and Why
| Outcome | Market | Model | Why |
|---|---|---|---|
| Above 52.4% | 99.0% | 99.8% | The Vercel AI Gateway report for September 2026 shows 71.5% open-weight token share, exceeding 52.4%. |
| Above 56.0% | 99.0% | 99.8% | The Vercel AI Gateway report for September 2026 shows 71.5% open-weight token share, exceeding 56.0%. |
| Above 74.0% | 4.0% | 0.5% | The Vercel AI Gateway report for September 2026 shows 71.5% open-weight token share, falling below 74.0%. |
| Above 57.8% | 90.0% | 99.8% | The Vercel AI Gateway report for September 2026 shows 71.5% open-weight token share, exceeding 57.8%. |
| Above 68.6% | 10.0% | 18.5% | The Vercel AI Gateway report for September 2026 shows 71.5% open-weight token share, supporting this threshold. |
Current Context
Public router data shows open-weight models dominating token volume. In September 2026, open-weight models processed 71.5% of Vercel AI Gateway tokens, representing 23% of estimated spend [1][2]. Open-weight spend on the gateway rose from about 5% in June to nearly one-quarter in September, a more than 70-fold increase since December 2025 [1]. OpenRouter production traffic on October 25, 2026, reported open-weight models at 79.6% of tokens versus 19.3% for closed models [3]. This specific router measure does not reflect economy-wide market share [3]. OpenRouter weekly volume data from October 2025 to October 2026 show open-weight tokens rising from 1.27 trillion to 120.95 trillion and closed-weight tokens from 3.60 trillion to 46.70 trillion. Open weights comprised approximately 74% of this increase [4]. Closed-weight volume grew 13-fold in this period, with its share recently fluctuating between 25% and 34% [4]. OpenRouter's data indicate open-weight token share surpassed 50% in April 2026, reached approximately 79% in early July, and stayed above 62% thereafter [4][5]. On one large public router, 368 trillion open-weight tokens were served in September 2026, up 57% daily from August; DeepSeek held 33.3% of this volume, GLM 20.8%, Hunyuan 19.4%, and MiMo 8.7% [6].
Gateway metrics offer limited view of overall market dynamics. Router and gateway token shares, while indicating rapid open-weight adoption, omit direct APIs, first-party applications, private deployments, and much enterprise self-hosting [7][8][9][10]. Earlier research found closed models dominated most tokens through 2025 [7][8]. Open models offer materially lower inference costs and flexibility [7][8][10]. The market displays a two-layer pattern: open weights dominate high-volume, price-sensitive inference on public routers, while closed frontier models retain disproportionate paid spend, enterprise preference, and capability/risk advantages [9][10]. Vercel's September data implies roughly 3.1 times as much open-weight token volume per dollar as the overall closed/open spending mix would suggest [1][9][10].
Open-weight share will likely remain strong on routers, but global share is uncertain. An October 17, 2026, scenario forecast anchors its Q4 2026 worldwide open-weight token share at 35%, explicitly labeling this value an assumption rather than an observed economy-wide statistic [11]. The forecast’s scenarios range from 31% to 39% by Q4 2027 and 16% to 59% by Q4 2035 [11]. Over the next 12–18 months, open-weight share is likely to remain a majority on developer-facing routers, potentially holding at 60%–80% [1][5][12][11]. A global all-channel share, however, cannot be inferred from these sources and may remain materially lower, as direct closed-provider traffic and enterprise deployments are undercounted [1][5][12][11]. Continued coexistence is the likely outcome, rather than total displacement [1][5][12][11].
Sources (12)
- 1Open weight takes 71.5% of tokens and 23% of spendvercel.com
- 2Vercel Index: Open-weight model use goes over 70% for the first time | heise onlineheise.de
- 3Open vs. Closed AI Models - Usage Share Tracked Daily | SmophyAIsmophy.ai
- 4Fast Fact: Closed-weight AI usage on OpenRouter rose 13-fold as its share of tokens fell - Global Advisors | Quantified Strategy Consultingglobaladvisors.biz
- 5The Token State of Play - Inferred | OpenRouteropenrouter.ai
- 6Open-Weight Token Report: October 2026 | American Computeamcompute.com
- 7The Emerging Market for Intelligence: Pricing, Supply, and Demand for LLMsnber.org
- 8AI open models have benefits. So why aren’t they more widely used? | MIT Sloanmitsloan.mit.edu
- 9WP-MI.EMS-01 v1.0 - Open Weights, Closed Frontiersai.sps.columbia.edu
- 10Open-weight AI: when the model becomes a utility - Everest Group Research Portaleverestgrp.com
- 11Open-weight share of tokens | AI Forecast Atlasaiforecastatlas.com
- 12Open-Weight Models Are Gaining Ground in Enterprise AI | Constellation Researchconstellationr.com
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: Above 60.0%
📉 October 26, 2026: 63.0pp drop
Price decreased from 68.0% to 5.0%
What happened: The reported 63.0 percentage point drop in the "Open vs. Closed Token Volume" prediction market for Oct. 26, 2026, is not supported by the provided evidence, which explicitly states "The phrase “drop 63.0pp” is not supported by the retrieved evidence" [1][2][3]. Instead, the underlying trend for open-weight models' share of AI token traffic was strongly upward, with Palanor reporting 56% on Oct. 18, 2026, a significant increase since Oct. 5, 2025 [2][4][5][6]. An X post from Social Capital on Oct. 6, 2026, showing open-weight token traffic rising to roughly 67%, was an identifiable social-media catalyst supporting this upward trend, coinciding with economic factors like much cheaper open-weight inference [7][5][6][8]. Therefore, no primary driver can be identified for the unsubstantiated 63.0 percentage point drop, making social media irrelevant to that specific, unconfirmed movement.
📈 October 23, 2026: 35.0pp spike
Price increased from 32.0% to 67.0%
What happened: The research does not identify a specific social media post or traditional news announcement that caused the 35.0 percentage point price spike in the prediction market on October 23, 2026 [9]. Prediction market digests for October 22 and 23, 2026, did not document a catalyst for such a sharp move in open-weight token share [9]. While Vercel's October 2026 production report, covering the September reporting period, announced open-weight models handling 71.5% of token volume [10], this indicates a sustained trend and was published prior to October 23, 2026 [11]. Earlier social media activity from August 2026 highlighted a growing open-weight share [12], representing a continuous narrative rather than an October 23 trigger. Therefore, social media activity was mostly noise or irrelevant to this specific price spike.
Outcome: Above 52.4%
📈 October 25, 2026: 10.0pp spike
Price increased from 75.0% to 85.0%
What happened:
The reported 10.0 percentage point spike in the prediction market cannot be verified from the available evidence [13]. However, around October 25, multiple independent trackers reported open-weight AI model usage significantly above the 52.4% threshold, with figures ranging from 60% to 79.6% of token volume [14][15][16][17]. A likely contributing factor was the reporting, described as an "edited Tech Twitter headline," that OpenRouter token demand had surged dramatically, with open-weight models handling about 72% of that demand [15]. This social media activity, appearing to coincide with the market date, would have reinforced existing data trends.
Social media, specifically the "edited Tech Twitter headline" regarding OpenRouter demand and open-weight model share, likely acted as a contributing accelerant by drawing attention to favorable underlying usage trends, though causality for the unverified spike is not established [15][13].
Outcome: Above 72.1%
📉 October 24, 2026: 16.0pp drop
Price decreased from 22.0% to 6.0%
What happened: The provided evidence does not substantiate a 16.0 percentage point drop for the "Above 72.1%" outcome in the "Open vs. Closed Token Volume" market on October 24, 2026 [18]. References to a 16.0 percentage point movement appear to relate to unrelated research rather than this specific prediction market [18]. Without confirmation of the alleged price movement, no primary driver can be identified for this specific event. Therefore, social media activity is irrelevant in this context.
Sources (18)
- 1Open vs. Closed Token Volume (open weights share) in Monday Odds & Predictions 2026kalshi.com
- 2Open vs. Closed Token Volume (open weights share) in Thursday Odds & Predictions 2026kalshi.com
- 3Open vs. Closed Token Volume (open weights share) in Wednesday Odds & Predictions 2026kalshi.com
- 4Open-Weight Models’ Share of AI Traffic — Chart — Palanorpalanor.com
- 5https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeordaily.dev
- 6Open weight models are dominating token share on Vercel's AI Gateway. | Ajay Y.linkedin.com
- 7Social Capital Open-weight AI model usage: 30.1K Views, 7.3x Its Typical Post | ViralChartviral-chart.com
- 8China Dominates Open Source AI Market with 60% Share Amid U.S. Struggles | Aju Pressajupress.com
- 9预测市场全景日报 | FLASH-FILLINGflash-filling.com
- 10Open weight takes 71.5% of tokens and 23% of spendvercel.com
- 11Open-weight models win hearts, minds, and walletstheregister.com
- 12This trend from vercel is so much reflective of what's happening at least in the non-enterprise world, curious to see a similar trend that can be reflextive of enterprise world too, expecting …linkedin.com
- 13Open weights AI market share (KXOPENSOURCESHARE): Kalshi odds history, order book & trades · TickFoundrytickfoundry.com
- 14Open vs. Closed AI Models - Usage Share Tracked Daily | SmophyAIsmophy.ai
- 15Open-weight models now handle 72% of OpenRouter token demand | Zeitgraphszeitgraphs.com
- 16Token Market Share by Lab — Diracdirac.run
- 17Token Demand Index — AI Inference Consumption & Effective Price | Bargobargo.ai
- 18Open weight models are 55% of our tokens and 15% of our spend: what the 78.4% Vercel number does and does not tell you | Requestyrequesty.ai
4. Market Data
Contract Snapshot
A 'Yes' resolution occurs if the market's stated 'Open vs. Closed Token Volume (open weights share)' on October 27, 2023, is above the specific percentage threshold for that contract. Conversely, a 'No' resolution occurs if it is at or below that threshold. The market begins on October 27, 2023, at 12:00 AM EDT, with a maximum payout date set for October 27, 2026.
Available Contracts
Market options and current pricing
| Outcome bucket | Yes (price) | No (price) | Last trade probability |
|---|---|---|---|
| Above 52.4% | $0.99 | $0.13 | 99% |
| Above 56.0% | $1.00 | $0.26 | 99% |
| Above 57.8% | $0.88 | $0.24 | 90% |
| Above 68.6% | $0.48 | $0.73 | 10% |
| Above 60.0% | $0.50 | $0.75 | 5% |
| Above 63.0% | $0.84 | $0.74 | 5% |
| Above 65.1% | $0.52 | $0.65 | 4% |
| Above 74.0% | $0.06 | $0.98 | 4% |
| Above 72.1% | $0.86 | $0.98 | 2% |
Market Discussion
Open-weight models appear to capture a significant majority of token volume, with reports indicating 71.5% in September 2026 [1] and 79.6% as of October 25, 2026, through 00:00 UTC October 26 [2]. Despite this high token usage, commentary suggests closed models retain a disproportionate share of spending [3][4]. Prediction markets reflect strong trader expectations for open-weight models to maintain a majority share, with recent contract resolutions and implied probabilities often falling in the high 60s [5][6][7][8].
Sources (8)
- 1Open weight takes 71.5% of tokens and 23% of spendvercel.com
- 2Open vs. Closed AI Models - Usage Share Tracked Daily | SmophyAIsmophy.ai
- 3Open-weight models win hearts, minds, and walletstheregister.com
- 4Chart of the Week: Open-Weight Models Win Tokens, Closed...galaxy.com
- 5Open vs. Closed Token Volume (open weights share) today Odds & Predictions 2026kalshi.com
- 6Open vs. Closed Token Volume (open weights share) in Thursday Odds & Predictions 2026kalshi.com
- 7Open vs. Closed Token Volume (open weights share) today Odds & Predictions 2026kalshi.com
- 8Open vs. Closed Token Volume (open weights share) in Wednesday Odds & Predictions 2026kalshi.com
5. Trust Index
“KXOPENSOURCESHARE 26OCT27 T60.0” made a sharp jump with almost no trading behind it.
Integrity risk· Thin-volume moves
How it adds up
Integrity80% of score79Good
Market integrity is low (69), but Integrity averages all three scores, so the other two pull it up. Only a critically low score would cap the total.
Trade quality20% of score58Caution
Includes the cost to trade: a $10,000 order can't be filled here because the order book is too thin.
Trust score75Good
Weighted blend with hard caps — a critically weak safety pillar, or a severe trading anomaly, caps the total regardless of the rest. Full methodology · About the Trust Index
Trust profile
Integrity4 screens run · 6 don't apply
6. How does the total cost of ownership for self-hosted open-weight models like Llama 3 compare to the API costs of closed-weight models like GPT-4o for high-volume enterprise tasks in Q3-Q4 2026?
| Self-hosted 70B Model Output Cost | $0.10 to $0.54 per million output tokens (Q3-Q4 2026) [1][2][3][4] |
|---|---|
| Proprietary API Output Cost (GPT-4o) | $10.00 per million output tokens [1][2][3][4] |
| Open-weight Model Market Share | Expected to exceed 57.8% (by October 26, 2026) [5][6][7] |
Self-hosting open-weight models offers substantial cost savings for high-volume enterprise tasks. In Q3-Q4 2026, the projected cost for self-hosted 70B models ranges from approximately $0.10 to $0.54 per million output tokens, varying with hardware utilization and model architecture [1][2][3][4]. This pricing significantly undercuts proprietary API services like GPT-4o, which are currently priced at $10.00 per million output tokens [1][2][3][4]. Enterprises can typically reach a break-even point for self-hosting at high volumes, estimated around 2 million tokens per day for optimized configurations [1][2][3][4].
However, the total cost of ownership for self-hosted models includes significant indirect expenditures. These include engineering time dedicated to infrastructure setup, ongoing maintenance, and monitoring, which can add 20-40% to the base GPU rental costs [1][8][4]. Consequently, self-hosting becomes a less attractive option for workloads characterized by low volume or bursty demand when compared to the pay-per-token model of API services [1][8][4].
Open-weight models are rapidly gaining significant market share by token volume. As of October 26, 2026, prediction markets indicate high confidence, specifically 90% at 57.8%, that the token volume share of open-weight models will exceed 57.8% on this date [5][6][7]. This strong trend reflects the competitive pricing and increasing adoption of open-weight alternatives across enterprise and developer environments [5][6][7].
Sources (8)
- 1LLM Inference Cost 2026: Cost per Million Tokenspacket.ai
- 2Llama 3.1 8B Inference Benchmarks 2026simplismart.ai
- 3The Real Cost of AI Inference in 2026: A Practical Breakdownblog.gpu.net
- 4AI Inference Cost Economics in 2026: GPU FinOps Playbookspheron.network
- 5Open vs. Closed Token Volume (open weights share) in Oct 26, 2026 | PredictPediapredictpedia.com
- 6Open vs. Closed Token Volume (open weights share) in Oct 26, 2026 Odds & Predictions - Public.compublic.com
- 7Top Open Source AI this week Prediction Market Results - Public.compublic.com
- 8Open-Source vs Proprietary LLM Inference Costgmicloud.ai
7. What evidence from Q3-Q4 2026 cloud provider earnings or direct API usage reports from OpenAI and Anthropic suggests that public router data overstates the true global market share of open-weight models?
| Open-Weight Models Token Share (Vercel AI Gateway) | 56% of tokens (August 2026) [1][2] |
|---|---|
| Anthropic Q3 2026 Revenue | $16.8B [3] |
| Anthropic July 2026 Run Rate | $65B [3][4][1] |
Public router data tends to overstate open-weight models' global market share. This occurs because it primarily measures developer-heavy, price-sensitive traffic and often counts requests rather than actual usage or economic value, while excluding substantial first-party and enterprise usage of closed models [1][2][5]. For example, OpenRouter only tracks routed developer traffic and reports requests, omitting significant first-party ChatGPT and Claude traffic and enterprise API usage from its rankings [1][2]. This methodological approach ultimately leads to request rankings that disproportionately favor cheap, high-volume models [2].
Closed models command a disproportionately larger share of market spending. Evidence from the Vercel AI Gateway in August 2026 demonstrated that open-weight models carried 56% of tokens but accounted for only 14% of total spending, whereas Anthropic alone captured 64% of the spending [1][2]. This indicates that open models dominate low-price volume but do not hold an equivalent economic share [1][2]. In contrast, closed frontier providers reported very large direct demand in Q3 2026. Anthropic achieved $16.8 billion in Q3 revenue, with its July run rate estimated near $65 billion, and OpenAI's August run rate exceeded $40 billion [3][4][1]. These substantial figures underscore significant first-party and enterprise consumption that remains uncaptured by public router datasets [3][4][1]. Anthropic's direct availability through major cloud providers such as AWS Bedrock, Google Vertex AI, and Microsoft Azure further supports the existence of this direct demand, which is invisible to router data [6][1].
Router data significantly overstates open-weight models' overall market presence. While there is no single authoritative global meter for open-versus-closed token volume, the available evidence strongly suggests that public router data, if extrapolated to represent all inference, likely overstates the global token share of open-weight models [7][5]. A precise corrected percentage for this overstatement is not currently established [7][5].
Sources (7)
- 1Does OpenAI still have a moat?newmarketpitch.com
- 2LLM Market Share 2026: ChatGPT, Gemini, Claude and Moreusecarly.com
- 3Anthropic Q3 revenue climbs to $16.8 billion: Reportfinance.yahoo.com
- 4October 8, 2026 - Nathan Benaich AIR STREET CAPITAL.stateof.ai
- 5Data - Authoritative AI Usage Data for Researchopenrouter.ai
- 6Anthropic expands Google and Broadcom compute dealanthropic.com
- 7Tokens Per Day · the AI inference demand indextokensperday.com
8. How might the potential releases of 'GPT-5' by OpenAI or a 'Llama 4' by Meta before October 2026 impact the token volume distribution between closed and open-weight models?
| Meta Llama 4 release | April 2025 (Scout and Maverick) [1][2] |
|---|---|
| OpenAI GPT-5.5 release | April 2026 [3][4][5] |
| Open-weight token share prediction | 99% above 52.4% by Oct. 26, 2026 [6][7][8] |
Initial release timings for major models are now outdated. The original premise concerning the first release of 'GPT-5' or 'Llama 4' by October 2026 has been superseded by new information. Meta introduced Llama 4 Scout and Maverick, which are open-weight Mixture-of-Experts (MoE) models, in April 2025 [1][2]. Subsequently, OpenAI announced GPT-5.5 in April 2026 and GPT-5.6 in July 2026, both characterized as closed frontier model families [3][4][5]. Consequently, the current discussion centers on the marginal effect of subsequent GPT-5.x releases rather than their initial launch timing [3][4][5].
New model releases will impact token volume distribution differently. A new closed GPT-5.x release would initially increase closed token volume, driven by factors such as quality, reliability, and established first-party or API distribution [3][4][2][5]. This could potentially reduce the open-weight market share, particularly when measured on hosted routers [9][10][6]. Conversely, an open-weight Llama release would directly contribute to increased open token volume, primarily by facilitating self-hosting, fine-tuning, and offering more cost-effective inference options [2][5]. However, Meta's shutdown of its hosted Llama API in 2026 suggests that the availability of downloadable weights does not automatically translate into hosted-router volume [2][5].
Open-weight models are highly predicted to dominate token share. Prediction markets strongly anticipate an open-weight majority by October 26, 2026, with reported odds of 99% for open-weight token share to exceed 52.4% and 99% for it to surpass 56.0% [6][7][8]. The base-case forecast indicates that the market will likely resolve above these thresholds, with a reasonable central estimate ranging from 60–65% open share [9][10][6]. A highly capable, fresh open-weight Llama release could potentially push this result toward or even above 68% [9][10][6]. Nevertheless, the dominant driver for open-weight token share by October 2026 is anticipated to be the broader low-cost open ecosystem, encompassing models from providers like DeepSeek, Qwen, Kimi, GLM, and MiniMax, rather than Llama 4 alone [9][11].
Sources (11)
- 1Evolution of Meta's LLaMA Models and Parameter-Efficient...arxiv.org
- 2Llama 4 Maverick 17B Instruct Usage, Cost & Rankopencode.ai
- 3GPT-5.6: Frontier intelligence that scales with your ambitionopenai.com
- 4Introducing GPT-5.5openai.com
- 5Meta Llama API Shutdown: Self-Hosting Is Now Official...spheron.network
- 6Open vs. Closed Token Volume (open weights share) in Oct 26, 2026 Odds & Predictions - Public.compublic.com
- 7Open vs Closed Token Volume (open… in Oct 26 Odds | Kalshipredictmarketcap.com
- 8Open vs. Closed Token Volume (open weights share) in Oct 26, 2026 | PredictPediapredictpedia.com
- 9State of AI 2025: 100T Token LLM Usage Studyopenrouter.ai
- 10The End of Best AI Model Wins era. Where Does the AI...uncoveralpha.com
- 11China's Open-Weight Takeover - by Chris Zeolidatagravity.dev
9. What are the primary data sources, such as the Vercel AI Gateway and OpenRouter reports, used to measure token volume, and what are their known limitations for estimating global market share as of October 2026?
| Prediction Market Resolution Date | Oct. 26, 2026 (Vercel AI Gateway) [1][2][3] |
|---|---|
| Open-weight models' share of Vercel gateway token volume | 56% (August 2026) [2][4][5] |
| OpenRouter data scope | billions of generations [6][7] |
Vercel AI Gateway and OpenRouter are primary sources for measuring token volume. Vercel's AI Gateway serves as the operative source for prediction market resolutions for October 26, 2026, measuring tokens routed through its gateway, encompassing input, output, reasoning, cached-input, and cache-creation tokens [1][2][3][8][5]. OpenRouter functions as a main complementary source, exposing top models by total token usage from its global platform. It achieves this by utilizing anonymized request-level metadata derived from billions of generations [9][6][7].
Estimating global market share faces significant coverage and ambiguity challenges as of October 2026. Neither the Vercel AI Gateway nor OpenRouter reports supply global token totals, leading to coverage bias and denominator ambiguity [1][2][9][6][5]. Vercel exclusively measures gateway-routed traffic, while OpenRouter's data reflects only its own user base, which is influenced by its model availability, pricing, and user preferences [1][2][9][6][5]. Furthermore, results can be distorted by factors such as provider mix, routing configurations, customer composition, workload mix, and differing tokenization or counting conventions across platforms [1][2][9][6][5].
Token share differs from revenue or model-value share, which is a crucial distinction. Open-weight models, for instance, could process a majority of tokens but capture a smaller fraction of overall spend due to varying token prices and workloads [2][4][6][5]. For a defensible October 2026 estimate, Vercel is to be treated as the direct resolution oracle, while OpenRouter serves as an independent directional cross-check [1][9][6][5].
Sources (9)
- 1AI Gateway Request Logsvercel.com
- 2AI Gateway production indexvercel.com
- 3Open vs. Closed Token Volume (open weights share) in Oct 26, 2026 Odds & Predictions - Public.compublic.com
- 4Open-weight models now handle a majority of tokens on...thenewstack.io
- 5Open-Weight AI Adoption Statistics & Chartslocalclaw.io
- 6State of AI 2025: 100T Token LLM Usage Studyopenrouter.ai
- 7An Empirical 100 Trillion Token Study with OpenRouterarxiv.org
- 8AI Gatewayvercel.com
- 9Data - Authoritative AI Usage Data for Researchopenrouter.ai
10. What 2026 data on enterprise adoption and spending patterns indicates continued strong demand for closed-weight models from providers like Anthropic and Google, despite the volume growth of open-weight alternatives?
| Anthropic Business Token Spending Share | 60.1% (Ramp, week ending October 11, 2026) [1] |
|---|---|
| Anthropic Total Spend Share | 61% (Vercel, June 2026) [2] |
| Open-Weight Models Spend Share | Under 9% (Vercel, late August 2026) [3][2] |
Enterprise AI spending continues to favor closed models despite open-weight volume growth. Enterprise AI adoption and spending accelerated significantly in 2026, with closed-weight models, particularly from Anthropic, retaining a dominant share of business spending despite substantial growth in open-weight token volume. Ramp's enterprise transaction data for the week ending October 11, 2026, indicated that Anthropic captured 60.1% of business token spending and 58.9% of token volume, alongside an adoption rate of 43.8% among Ramp businesses [1]. Vercel's production-gateway data from June 2026 similarly showed open-weight models processing 29% of tokens but accounting for less than 4% of spending, while Anthropic processed 32% of tokens and captured 61% of total spend [2]. By late August 2026, open-weight models reportedly reached approximately 62% of Vercel AI Gateway token volume, yet captured under 9% of spending, with closed models retaining roughly 61%-65% of total spend [3][2].
Task segmentation and enterprise needs drive distinct usage patterns for models. This pattern indicates that open-weight models are favored for cheap, high-volume tasks such as classification, extraction, summarization, routing, and formatting [3][2]. Conversely, closed frontier models are concentrated in high-value reasoning, coding, and agentic work [3][2]. Enterprise survey evidence supports this task-segmented model estate, suggesting it is not a wholesale replacement of closed models [4]. While some firms utilized self-hosted open-weight models for cost efficiency, other surveyed firms cited quality, security, reliability, and support as constraints for open-weight adoption [4]. Overall, 2026 data supports continued strong enterprise demand for closed-weight models in economic terms—especially spend and high-value workflows—even if open-weight alternatives dominate or approach dominance in raw token volume, signifying that open-weight volume growth should not be interpreted as a collapse in closed-model demand [1][2][5].
11. What Could Change the Odds
Key Catalysts
The relevant prediction market for Vercel AI Gateway's October 26 open-weights percentage resolves against the scheduled reference/update at 10:00 AM ET on October 27, 2026 (2:00 PM UTC) [1][2][3][4][5]. Recent displayed contracts imply near-certainty above 52.4% and 56.0%, though higher thresholds carry materially less certainty [1][2][3][4][5]. The latest finalized Vercel AI Gateway data, from September 2026, showed open-weight models processed 71.5% of tokens but represented 23% of estimated spend; open-weight spend share increased from approximately 5% in June [1][2]. The central market structure dictates that token share does not equal revenue share, as low-cost open models attract token-heavy workloads like coding and agentic tasks, while closed frontier models retain disproportionate spend [1][2][6][7]. DeepSeek-V4.1-Flash handled nearly half of September tokens but only 2.6% of spend [1][2][6][7].
Bullish catalysts for open-weight share include continued price declines, rapid provider availability, Chinese model releases/repricing, and growth in token-intensive agentic workloads [1][8][9][10][7]. Conversely, bearish catalysts for open-weight share include a rebound in premium closed-model quality, proprietary product integration, enterprise governance preferences, or a change in Vercel's model classification or methodology [1][8][9][10][7].
Key Dates & Catalysts
- Expiration: November 03, 2026
- Closes: October 27, 2026
Sources (10)
- 1Open weight takes 71.5% of tokens and 23% of spendvercel.com
- 2Vercel Index: Open-weight model use goes over 70% for the first time | heise onlineheise.de
- 3Open vs. Closed Token Volume (open weights share) in Oct 26, 2026 Odds & Predictions - Public.compublic.com
- 4Open vs. Closed Token Volume (open weights share) in Saturday Odds & Predictions 2026kalshi.com
- 5Open vs. Closed Token Volume (open weights share) in Thursday Odds & Predictions 2026kalshi.com
- 6Open weight models are 55% of our tokens and 15% of our spend: what the 78.4% Vercel number does and does not tell you | Requestyrequesty.ai
- 7Token Demand Index — AI Inference Consumption & Effective Price | Bargobargo.ai
- 8The Open-Weight Token Report | American Computeamcompute.com
- 9The State of Open Source AI — v1.1 · September 2026stateofopensource.ai
- 10State of Open Source AI 2026stateofopensource.ai
13. Historical Resolutions
Historical Resolutions: 20 markets in this series
Outcomes: 19 resolved YES, 1 resolved NO
Recent resolutions:
- KXOPENSOURCESHARE-26OCT25-T73.0: YES (Oct 25, 2026)
- KXOPENSOURCESHARE-26OCT25-T70.1: YES (Oct 25, 2026)
- KXOPENSOURCESHARE-26OCT25-T69.1: YES (Oct 25, 2026)
- KXOPENSOURCESHARE-26OCT25-T65.4: YES (Oct 25, 2026)
- KXOPENSOURCESHARE-26OCT25-T63.2: YES (Oct 25, 2026)
