Short Answer

Despite a September 19 snapshot from the Vercel AI Gateway reportedly showing open-weight token volume at 78.4%, the market considers it unlikely that open source market share will be above 70% by September 28, 2026 (29.0%).

1. Market Behavior & Drivers

No specific development coincides with the market's 70 percentage point drop on September 28. The price collapsed from 97.0% to 27.0% on a volume of only two contracts. This suggests the move was driven by low liquidity or a specific participant's re-rating rather than a broad reaction to new fundamental information. Available research did not identify any news or social media narrative that would justify the repricing.
The negative price action is inconsistent with available usage data. Open-weight models are reported to have processed 56% of tokens on Vercel AI Gateway in August, up from 13% in April. On the OpenRouter platform, their traffic share exceeded 50% in September. A separate, related prediction market implies an over 84% probability of open-weight token share crossing a similar threshold, pointing to a significant divergence in expectations between the two contracts.
  • Open-weight token share appears above 70% based on recent Vercel data.
  • A September 19 Vercel AI Gateway snapshot reportedly showed 78.4% open-weight volume.

Who Wins and Why

Outcome Market Model Why
Above 67.5% 23.0% 37.7% The Vercel AI Gateway's September 19 snapshot reported open-weight token volume at 78.4%.
Above 82.5% 1.0% 0.8% Open-weight token volume reportedly showed strong and continued growth by September 19.
Above 87.5% 1.0% 0.8% Open-weight token volume reportedly showed strong and continued growth by September 19.
Above 85% 1.0% 0.8% Open-weight token volume reportedly showed strong and continued growth by September 19.
Above 80% 1.0% 1.8% Open-weight token volume reportedly showed strong and continued growth by September 19.

Current Context

Open-weight AI models capture increasing token share, particularly in inference. As of September 28, 2026, open-weight models processed 56% of tokens on Vercel AI Gateway in August, a rise from 13% in April, while representing only 14% of associated spending [1]. On other routed API platforms, open weights exceeded 50% of traffic on OpenRouter in September, though open-model API revenue remained below 20% of platform revenue [2]. A Kalshi-linked market, settling October 1 on September 30 data, implies an above 84% probability of open-weight token share crossing a 66%/69% threshold, but its threshold variants require cautious interpretation [3]. Constellation Research cited a September 2026 survey showing open-weight models at 34% of enterprise AI token usage, up from 23% a year prior [4]. Expert analysis converges on a two-tier market where open models secure high-volume, cost-sensitive inference, while closed models maintain disproportionate premium spend and frontier positioning [5]. Chinese open-weight models are gaining competitive gravity, exceeding 80% of OpenRouter open-model usage in September 2026 and garnering mentions in about 30% of sampled research papers for Qwen versus 21% for Llama [6]. Cumulative open language model downloads reached 2.04 billion by March 2026, a 6x year-over-year increase, with Chinese models accounting for 1.15 billion downloads [7].
Enterprise open-weight model adoption is progressing from pilot to production. Production adoption rose from 31% in July to 42% in September, with 43% of enterprises still piloting models [4]. Beyond AI, open-source software remains fundamental enterprise infrastructure, with reports emphasizing mission-critical use, security, compliance, digital sovereignty, and cloud-native deployment [8]. Industry audits cited in a 2026 study found open-source code in approximately 96% of commercial codebases [9]. AI-assisted development is rapidly increasing within open source; AI-attributed commits across over 180 million repositories rose from 1.6% of non-bot activity in December 2025 to 6.7% by March 2026 [10]. Specialized ecosystems, such as quantum software on GitHub, show significant collaboration, with over 1,500 repositories and 27,000 contributors in a June 2026 snapshot [11].
Open-source software remains foundational, but market share metrics vary by domain. No authoritative source provides a single global "open-source market share" percentage as of September 28, 2026; defensible reporting relies on domain-specific indicators [11]. A commercial forecast estimates the global open-source software market at $56.57 billion in 2026, projected to reach $95.38 billion by 2030, but this reflects revenue, not usage share [12]. Licensing structures present a market constraint; a 2026 study found 83% of over 131 million projects lacked a formal license, while MIT represented 65% of licensed projects [13]. Open-source data-infrastructure sustainability is under pressure, with license and sustainability events rising approximately 55% over eight years, associated with cloud-managed-service commoditization [14]. Near-term, open-weight token share is likely to continue rising through late 2026, but a complete takeover of total AI revenue is unlikely due to premium closed models retaining spending share and first-party usage exclusions from some platform datasets [5]. Monetization and maintainer sustainability represent principal risks [11].
Sources (14)
  1. 1Open weight models process majority of AI tokens for the first time - TechCentral.ietechcentral.ie
  2. 2Notes on Open Source - Market Sentimentmarketsentiment.co
  3. 3Open Source market share on Sep 30th - Kalshi Odds | CoinRithmcoinrithm.com
  4. 4Evidence Shows Enterprises Use of Open Weight Models is Mainstream | Constellation Researchconstellationr.com
  5. 5State of Open Source AI 2026stateofopensource.ai
  6. 6The current balance of power in open models | LavX Newsnews.lavx.hu
  7. 7The ATOM Report: Measuring the Open Language Model Ecosystemarxiv.org
  8. 8Open-Source Software Report 2026: 4,496 Projects Analyzeddev.co
  9. 9Project Life Cycles in Open-Source Softwarearxiv.org
  10. 10Detecting AI Coding Agents in Open Source: A Validated Multi-Method Census of 180 Million Repositoriesarxiv.org
  11. 11Entangle: Uncovering Collaboration in the GitHub Quantum Software Ecosystemarxiv.org
  12. 12Open Source Software Market Size, Trends Report 2026-2030thebusinessresearchcompany.com
  13. 13[2606.23445v1] The Prevalence and Impact of Licenses in Open Software Projectsarxiv.org
  14. 14Why Memory Components Fail: Eight Years of License and Sustainability Events in Open-Source Data Infrastructurearxiv.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: Above 67.5%

📉 September 28, 2026: 70.0pp drop

Price decreased from 97.0% to 27.0%

What happened: The provided web research does not establish a specific primary driver for the reported 70.0 percentage point drop in the "Open Source market share on Sep 28th" Kalshi market. No social media activity, including posts from key figures or viral narratives, was identified to have caused or coincided with such a significant price movement around September 28, 2026. While news of open-source initiatives from Nvidia and Meta were noted, the sources explicitly state these do not prove a direct cause for the Kalshi market repricing [1][2]. Therefore, based on the available information, social media activity was irrelevant as a primary driver or contributing accelerant for this market change.

Outcome: Above 70%

📈 September 27, 2026: 62.0pp spike

Price increased from 2.0% to 64.0%

What happened: While the provided evidence does not explicitly establish a 62.0 percentage-point spike on September 27, 2026 [3][4][5], the primary driver for a positive price movement for the "Above 70%" outcome appears to be the strong performance of open-weight models. Vercel AI Gateway data reported an open-weight token share of 78.4% on September 18–19, significantly exceeding the market's threshold, with the market explicitly resolving based on Vercel's chart [6][7][8][9][10][11]. Although this influential data was reported earlier, the prediction market likely fully priced in this information as the September 29 resolution date neared. No specific social media activity, such as posts from key figures or viral narratives, was identified as a direct catalyst for a September 27 spike, making it mostly noise or irrelevant in this instance.
Sources (11)
  1. 1Nvidia launches OpenShell, an open-source runtime for securing autonomous AI agentscryptobriefing.com
  2. 2Meta AI's Muse Glimmer: 30B Open-Source Model for Consumer Gfuturesignalnews.com
  3. 3Economies of Open Intelligence: Tracing Power & Participation in the Model Ecosystemarxiv.org
  4. 4https://arxiv.org/pdf/2602.17131arxiv.org
  5. 5https://arxiv.org/pdf/2509.20010arxiv.org
  6. 6Open Source market share on Sep 28th - Kalshi Odds | CoinRithmcoinrithm.com
  7. 7Open Source market share on Sep 28th - Probabilidades na Kalshi | CoinRithmcoinrithm.com
  8. 8Open 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
  9. 9Share Of Closed Models Has Fallen From Around 70% To 21% In The Last 3 Months: Vercel Dataofficechai.com
  10. 10Vercel AI Gateway shows open models dominating token volume over closed ones - Ranzware – Innovation, Reviews & Trendsranzware.com
  11. 11Open-Weight AI Models Seize Token Lead, but Proprietary Systems Still Make the Moneytechstrong.ai

4. Market Data

Contract Snapshot

A "YES" resolution occurs if the open-source market share is determined to be above 67.5% on September 28th. Conversely, a "NO" resolution occurs if the open-source market share is 67.5% or below on that date. The market's outcome is based on data from September 28th, with a maximum payout date of September 29, 2026. No specific settlement conditions beyond the percentage threshold are mentioned.

Available Contracts

Market options and current pricing

Outcome bucket Yes (price) No (price) Last trade probability
Above 70% $0.23 $0.93 29%
Above 67.5% $0.23 $0.82 23%
Above 75% $0.07 $0.99 3%
Above 77.5% $0.02 $0.99 1%
Above 80% $0.02 $0.99 1%
Above 82.5% $0.02 $1.00 1%
Above 85% $0.01 $1.00 1%
Above 87.5% $0.02 $1.00 1%
Above 72.5% $0.22 $0.97 0%

Market Discussion

The "Open Source market share on Sep 28th" market on a specific prediction platform prices the outcome "Above 67.5%" at a 97% implied probability, scheduled to resolve on September 29, 2026, using data from a specific AI Gateway's token volume chart for September 28 [1]. This market references open-weight token share, which has shown historical growth from May 29 to August 26, 2026, and is supported by broader bullish narratives for open-weight AI, especially from Chinese models [1]. However, a key distinction exists between open-weight and truly open-source AI, with transparency declining in the latter; experts also caution that prediction market prices may reflect localized information and biases rather than definitive forecasts [2].

Sources (2)
  1. 1Open Source market share on Sep 28th - Kalshi Odds | CoinRithmcoinrithm.com
  2. 2Market Beliefs about Open vs. Closed AIarxiv.org

5. Trust Index

Octagon Trust Index Kalshi 72 Good

“KXOPENSOURCESHARE 26SEP29 T70” made a sharp jump with almost no trading behind it.

Integrity risk· Thin-volume moves

How it adds up
Integrity80% of score77Good

Market integrity is low (63), 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 score52Caution

Includes the cost to trade: a $10,000 order can't be filled here because the order book is too thin.

Trust score72Good

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
Market integrity
63Caution

Outcome can be influenced

Thin-volume movesTrade signals · high confidence

“KXOPENSOURCESHARE 26SEP29 T70” made a sharp jump with almost no trading behind it.

Large tradesTrade signals · medium confidence

Largest single print on KXOPENSOURCESHARE-26SEP29-T67.5 carried 48% of its $413 tape (window is 38% of 24h flow)

Screened clean
Size patternsNo distribution read Clear
2 screens not run
Outcome control · doesn't apply to this market type
Thin-leg conviction · doesn't apply to this market type
Info fairness
80Good

Structural prior — no screens apply

No suspicious activity on any screen

Screened clean
Information exposureNo concentrated group can know the outcome materially early. Clear
4 screens not run
Move vs. catalyst · doesn't apply to this market type
Longshot repricing · doesn't apply to this market type
Pre-broadcast certainty · doesn't apply to this market type
Cross-venue timing · doesn't apply to this market type
Resolution quality
92Strong

Resolution mechanism is dependable

No suspicious activity on any screen

Resolution rules read: Clear

Liquidity
66Tradeable

Fills fine at modest size

7 contracts · worst first · 2 below the volume floor — not scored

Above 77.5%49
Thin · low confidence

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

Above 85%50
Thin · low confidence

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

Above 87.5%51
Thin · low confidence

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

Above 82.5%52
Thin · low confidence

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

Above 80%53
Thin · low confidence

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

Above 75%59
Thin · low confidence

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

Above 67.5%80
Tradeable · low confidence

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

Book spread 26% vs realized 2%
Move quality
75Confirmed

Moves came with real volume

7 contracts · worst first · 2 below the volume floor — not scored

Above 67.5%69
Mostly confirmed · high confidence

Volume confirms the move over 6h

Above 77.5%78
Confirmed · high confidence

Volume confirms the move over 6h

Above 75%80
Confirmed · high confidence

Volume confirms the move over 6h

Above 80%82
Confirmed · high confidence

Volume confirms the move over 6h

Above 87.5%85
Confirmed · high confidence

Volume confirms the move over 6h

Above 85%87
Confirmed · high confidence

Volume confirms the move over 6h

Above 82.5%89
Confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100Clear

Counts toward Resolution quality, not here

7 contracts · worst first · 2 below the volume floor — not scored

Above 67.5%100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 75%100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 77.5%100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 80%100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 82.5%100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 85%100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 87.5%100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Trade quality
52Caution

Order book is thin

Tradable, but expect execution cost beyond small sizes.

Outcome quality 67 (notional-weighted across 7 scored legs)

$100 order: 26.5c slippage (~129% of a 20.5c contract)

$1,000 order: 33.3c slippage (~162% of a 20.5c contract)

$10,000 order: book too thin to fill

2 below the volume floor — not scored

Above 67.5%Primary80Market scoreLiquidity80Move quality69Rule clarity100
Why 80
Liquidity
80 · Tradeable · low confidence

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

Book spread 26% vs realized 2%
Move quality
69 · Mostly confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 75%59Market scoreLiquidity59Move quality80Rule clarity100
Why 59
Liquidity
59 · Thin · low confidence

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

Move quality
80 · Confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 77.5%49Market scoreLiquidity49Move quality78Rule clarity100
Why 49
Liquidity
49 · Thin · low confidence

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

Move quality
78 · Confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 80%53Market scoreLiquidity53Move quality82Rule clarity100
Why 53
Liquidity
53 · Thin · low confidence

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

Move quality
82 · Confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 82.5%52Market scoreLiquidity52Move quality89Rule clarity100
Why 52
Liquidity
52 · Thin · low confidence

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

Move quality
89 · Confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 85%50Market scoreLiquidity50Move quality87Rule clarity100
Why 50
Liquidity
50 · Thin · low confidence

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

Move quality
87 · Confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Above 87.5%51Market scoreLiquidity51Move quality85Rule clarity100
Why 51
Liquidity
51 · Thin · low confidence

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

Move quality
85 · Confirmed · high confidence

Volume confirms the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear

underwriting_v1.4 · computed Sep 28, 2026

6. What upcoming flagship model releases or licensing changes from labs like Meta and Alibaba could significantly alter the open-source token share by September 2026?

Open-weight share by mid-2026above 67.5% [1][2][3][4][5]
Meta Muse Glimmer releaseAugust under Apache 2.0 [6][7][8]
Alibaba Qwen3.8 models releaseAugust 2026 under Apache 2.0 [9][10]
Open-weight models are projected to maintain a significant market share. Market odds indicate that open-weight models will retain over 67.5% of routed usage by mid-2026 [1][2][3][4][5]. This share could be significantly influenced by September 2026 through upcoming developments from Meta, specifically Muse Glimmer, and potential licensing decisions for Alibaba's Qwen4.
Meta’s recent releases could boost open-source model adoption. Meta released Muse Glimmer, a 30-billion-parameter multimodal agent model, in August 2026 under the permissive Apache 2.0 license [6][7][8]. This represents a shift from Meta's previous Llama Community License. While open weights were promised for Muse Spark 1.2, Spark 1.3 remained closed or API-oriented [11]. The next-generation Llama, codenamed Avocado, is anticipated in 2027, meaning a Llama 5 flagship model is not expected by September 2026 [12].
Alibaba’s future licensing decisions will shape its market impact. Alibaba contributed to open-source availability by releasing Qwen3.8-27B and Qwen3.8-2.4T-A95B weights in August 2026 under Apache 2.0 [9][10]. However, the licensing for Alibaba's next major model, Qwen4, which was still in training as of late September, remains unconfirmed for a downloadable release [13][14]. Although an early open-weight preview of the Qwen4 architecture, Qwen3.8-Flash-Next, was released in August, Qwen4's ultimate licensing status (Apache 2.0, restricted, or API-only) is unresolved [13][14][15]. Prior plans for Qwen3.8-Max had included revenue-sharing obligations for large commercial deployments [13][14][15][5].
Sources (15)
  1. 1State of Open Source AI 2026stateofopensource.ai
  2. 2The State of Open Source AI — v1, July 2026stateofopensource.ai
  3. 3Open-Source LLMs Overtook Proprietary in 2026capitalandcompute.net
  4. 4Open Source market share on Sep 28th - Kalshi Odds | CoinRithmcoinrithm.com
  5. 5The Great Model Migration — CodeSOTAort.fabryka.ai
  6. 6Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device | Meta AI Researchresearch.meta.ai
  7. 7Meta returns to open source with Muse Glimmer, an Apache 2.0 licensed 30B parameter AI model optimized for agents — available now | VentureBeatventurebeat.com
  8. 8Meta is back with Muse Glimmer: local, agentic, multimodal, and open sourcehuggingface.co
  9. 9Alibaba Unveils Qwen3.8-27B and Releases Weights of Qwen3.8 Flagship Model - Alibaba Cloud Communityalibabacloud.com
  10. 10Alibaba Qwen3.8-27B debuts with Apache 2.0 open-weight license – GenAI Dailygenaidaily.com
  11. 11Meta’s Next AI Models: Open Source Versions Are Comingweb.quintessentialtech.com
  12. 12Llama 5 Still Isn't Out — Meta's Real 2026 Release Is Muse Glimmer 30B (August 2026) — LLM Checkllmcheck.net
  13. 13Alibaba’s Qwen to open-source Qwen3.8-Flash-Next, previewing Qwen4 architecture · TechNodetechnode.com
  14. 14Qwen 4: Release Date, Leaks, and the Architecture Alibaba Already Shipped | CellCogcellcog.ai
  15. 15Alibaba prepares revenue-sharing terms for its next Qwen model – GenAI Dailygenaidaily.com

7. How does the API usage of leading Chinese open-weight models, such as Alibaba's Qwen, compare to Meta's Llama series on major routing platforms through mid-2026?

Qwen OpenRouter Request Share6.4% (week of Sep 21, 2026) [1][2]
Qwen OpenRouter Token Share12.7% (mid-2026) [3][4]
Meta/Llama OpenRouter Token Share0.0% (mid-2026) [3][4]
Alibaba's Qwen model significantly surpasses Meta's Llama series in API usage. On major routing platforms through mid-2026, Qwen's API usage materially exceeded that of Meta's Llama series, with the disparity likely several-fold and potentially an order of magnitude [1][3]. For the week starting September 21, 2026, OpenRouter's request-based author ranking reported Qwen at 6.4%, while the published top authors did not include any Meta or Llama entry [1][2]. Stronger token-based evidence from OpenRouter in mid-2026 further indicated Qwen's significant lead, accounting for 12.7% of current tracked tokens compared to Meta-Llama at 0.0% [3][4]. This data also noted Qwen's increase from 2.2% a year prior, concurrent with Meta-Llama's decline from 5.6% [3][4]. Secondary analyses generally estimate Qwen's share near 13%–14% of total volume in mid-2026, whereas Meta/Llama is frequently reported below 1% in July analyses [5][6][7][8].
Chinese open-weight models collectively lead in routing platform token share. OpenRouter's June 2026 analysis indicates that Chinese open-weight models collectively surpassed American models in token share by early June [9][2]. Across various secondary analyses of OpenRouter's mid-2026 data, Chinese-origin models are estimated to account for approximately 45%–61% of routed token volume [5][6][7][8]. It is important to note that OpenRouter explicitly cautions its request-based chart measures text requests, not tokens, users, revenue, or global market share [1][2]. Therefore, the most defensible interpretation of these findings for prediction markets through September 28, 2026, is platform-routed usage or share, rather than global AI market share [1][3]. This comparison specifically focuses on usage within routing platforms, which tend to favor economical and capable open-weight models, such as Qwen and Llama, for high-volume production workloads [10][9][11].
Sources (11)
  1. 1LLM Rankings | OpenRouteropenrouter.ai
  2. 2DeepSeek V4 Is Earning Agentic Token Share — OpenRouter Blogopenrouter.ai
  3. 3OpenRouter Model Author Token Share by Monththelatent.co
  4. 4OpenRouter Market Trends: One Year of LLM Inference | CodeSOTA | CodeSOTAcodesota.com
  5. 5Chinese AI Models Q2 2026: 10-Provider Landscape Reportdigitalapplied.com
  6. 6OpenRouter Model Usage Rankings 2026: Real Token Share | Presenc AIpresenc.ai
  7. 7Chinese AI Models Now Power Nearly Half of US Enterprise Tokens | metir Blogmetirai.com
  8. 8Chinese LLMs Take Top Five Spots On OpenRouter - Dataconomydataconomy.com
  9. 9The Open Weight Models that Matter: June 2026 — OpenRouter Blogopenrouter.ai
  10. 1030+ Open Source LLM Statistics & Trends (2026) — OpenLLMStackopenllmstack.com
  11. 11Lowest-Cost LLM Inference: The Complete OpenRouter Guide — OpenRouter Blogopenrouter.ai

8. According to recent enterprise surveys from firms like Constellation Research, what is the reported production adoption rate of open-source AI versus closed-source AI?

Open model production adoption51% (May 2026 SlashData/Mozilla) [1][2]
Closed model production adoption63% (May 2026 SlashData/Mozilla) [1][2]
Open-weight model production in enterprises42% (September 2026 Enterprise Technology Research) [3][4]
Closed-source AI models generally show higher production adoption rates. A May 2026 survey conducted by SlashData/Mozilla, involving 1,494 developers, reported that 51% of open-model users were running these models in production, in contrast to 63% for closed models. This indicates a 12-percentage-point production adoption gap favoring closed models [1][2]. For large enterprises specifically, the production adoption rates were approximately 57% for open models and 73% for closed models, maintaining a similar gap [1][2].
Open-weight models are increasingly adopted and gaining enterprise token share. A separate September 2026 Enterprise Technology Research (ETR) survey of 200 enterprises found that 42% of respondents had open-weight models in production, with an additional 43% still piloting them [3][4]. This survey also highlighted that open-weight models accounted for 34% of enterprise AI token usage, marking an increase from 23% a year prior, and are anticipated to reach 41% within the subsequent 12 months [3][4]. While other surveys addressed deployment mix and token volume, they did not offer direct comparisons of production adoption rates between open-source and proprietary models [5][6][7].
Sources (7)
  1. 1State of Open Source AI 2026stateofopensource.ai
  2. 2Open models: 79% adoption, 12-point production gapslashdata.co
  3. 3Evidence Shows Enterprises Use of Open Weight Models is Mainstream | Constellation Researchconstellationr.com
  4. 4Open-Weight AI Models Expected to Capture 41% of Enterprise Token Usagetechstrong.ai
  5. 5Enterprise AI Adoption Trends 2026 | Halkwinds Researchhalkwinds.com
  6. 6Open-Weight Models Are Gaining Ground in Enterprise AI | Constellation Researchconstellationr.com
  7. 7Open-Weight Models Are Gaining Ground in Enterprise AI | Constellation Researchconstellationr.com

9. Which API gateways, such as Vercel and OpenRouter, provide the most consistent public data on the token and revenue share of open-weight models for 2025-2026?

August 2026 Open-Weight Token Share56% (Vercel AI Gateway) [1]
August 2026 Open-Weight Spend Share14% (Vercel AI Gateway) [1]
May-Sep 2025 Open-Weight Revenue Share~4% (OpenRouter) [2]
Vercel AI Gateway offers the most consistent public data for open-weight models. It is identified as the most consistent public source for data on open-weight models' token share and estimated spend share for 2025-2026, with OpenRouter serving as a valuable complementary source for usage data [3]. The available information indicates that no gateway consistently publishes a comprehensive global revenue share for open-weight models, and reported figures reflect routed traffic rather than total market share [3].
Vercel's data shows significant growth in open-weight model adoption. Its September 2026 report indicates that open-weight models constituted 56% of gateway tokens and 14% of estimated spend in August 2026 [1]. The token share for open-weight models on Vercel AI Gateway demonstrated significant growth, rising from 7% in December 2025 to 13% in April 2026, 36% in July, and 56% in August. A daily snapshot on September 19 reportedly reached 78.4% [4]. Vercel also provides daily model/lab datasets through a CC BY 4.0 leaderboard-export API, which includes requests, tokens, and spend metrics [5].
OpenRouter provides valuable live usage data, but lacks spend and revenue. It offers public live usage data, including token-processed rankings and weekly author request shares, and its API documentation defines machine-readable sampled token_share and usage_share fields [6]. While useful for usage insights, OpenRouter's public rankings do not measure spend or revenue share [6]. During its last publicly described revenue window (May-September 2025), open-weight models accounted for approximately 20% of model-layer usage but only about 4% of revenue on OpenRouter [2]. Third-party trackers can assist in archiving and aggregating OpenRouter data but are derived, not primary, sources [7].
Sources (7)
  1. 1Open-weight models take 56% of token volume, Astra doubles Fable 5.1 spendvercel.com
  2. 2Open Models Are Winning AI Traffic, but Not the Moneyhackernoon.com
  3. 3AI Gateway production indexvercel.com
  4. 4Open-weight models account for 56 percent of monthly token throughput on Vercel AI Gatewaydigitaltoday.co.kr
  5. 5Access and share AI Gateway leaderboard datavercel.com
  6. 6LLM Rankings | OpenRouteropenrouter.ai
  7. 7AI Model Market Report — Token Share by Origin | Zingzing.co.uk

10. What is the total cost of ownership (TCO) difference for an enterprise deploying a top-tier open-source model versus using a flagship closed-source API from OpenAI or Anthropic in 2026?

TCO Crossover Token Volume10 million and 100 million tokens per month [1][2][3][4][5][6]
Self-hosting Savings40–90% [1][2][3][4][5][6]
Annual MLOps Engineering Cost$300k–$600k annually [1][3][4][5][7][8]
Enterprise LLM TCO depends on monthly token volume. For enterprises, the total cost of ownership (TCO) when choosing between deploying open-source models and utilizing flagship closed-source APIs typically exhibits a crossover point in token volume, falling between 10 million and 100 million tokens per month [1][2][3][4][5][6]. Below this specific volume, closed-source APIs are generally more cost-effective because they eliminate upfront infrastructure investments and ongoing maintenance costs [1][2][3][4][5][6].
High token volumes favor self-hosting despite significant fixed costs. Above the identified token volume, self-hosting open-source models can lead to significant savings, often ranging from 40–90%, when compared to using proprietary APIs [1][2][3][4][5][6]. However, self-hosting involves substantial fixed costs, including capital expenditure for GPUs or cloud rental, electricity, and dedicated engineering labor, which is frequently estimated at $300,000–$600,000 annually for MLOps [1][3][4][5][7][8]. These fixed costs must be carefully balanced against the linear, variable operating expenses associated with proprietary API consumption [1][3][4][5][7][8].
Strategic needs dictate choice between open-source and proprietary LLMs. Enterprises often prioritize proprietary APIs for advantages such as rapid iteration capabilities, access to state-of-the-art frontier reasoning models, and simplified operational management [1][3][4][6]. Conversely, self-hosting is typically adopted for high-volume, predictable workloads, stringent data residency requirements, and the need for deep model customization [1][3][4][6].
Sources (8)
  1. 1The Cost of Cloud LLM APIs vs Local Inference: A TCO Analysis | Cognisoccognisoc.com
  2. 2Open Source vs Proprietary LLMs: The Real Cost Breakdown — Kael Researchkaelresearch.com
  3. 3Open-Source vs Commercial LLMs: The Complete Guide (2026)sitepoint.com
  4. 4Open Source vs Closed LLMs in Enterprise: A Total Costcallsphere.ai
  5. 5Open-Source vs Proprietary LLMs: TCO Calculator | AICloudITaicloudit.com
  6. 6Open Source LLMs in 2026: Pricing, Hosting & How to Choosebuildtolaunch.substack.com
  7. 7A Cost-Benefit Analysis of On-Premise Large Language...arxiv.org
  8. 8LLM Inference On-Premise vs GPU Cloud: 2026 Cost and...spheron.network

11. What Could Change the Odds

Key Catalysts

The Kalshi contract "Open Source market share on Sep 28th" settles on September 29, 2026, based on Vercel AI Gateway's Open-vs-Closed Token Volume chart [1]. The key threshold for settlement requires open-weight share above 67.5% [1]. Current odds snapshot prices this outcome at 97% implied probability, though a base-case estimate projects a 70%–80% probability of clearing 67.5% [1]. This estimate is deliberately below the market price because the cited 56% August figure and roughly one-third broader figure do not independently prove a daily September 28 reading above 67.5%; the market itself and recent trend provide the stronger signal [1][2][3]. The measurement date is September 28, 2026, with the benchmark expected to update at 10:00 AM ET on September 29, 2026 [1][2].
A strong bullish catalyst for open-weight adoption is evident in Vercel AI Gateway data, which reported open-weight models at 56% of token volume in August 2026, a significant rise from 13% in April [2]. The State of Open Source AI report notes open models power roughly one-third of tokens, with the next four leading models behind the closed frontier being open [3]. Prediction market trends also lean bullish: the Sep 26 open-source contract priced clearing 69% at 99%, and Sep 28 weekly contracts showed 97% for DeepSeek clearing 23%, serving as momentum indicators [4][5]. The economic argument favors open models, which processed 56% of tokens with only 14% of AI Gateway spending, indicating a low-cost/high-volume advantage [2][3]. Broader open-source software revenue is forecast to grow from $56.57B in 2026 to $95.38B in 2030, a 14% CAGR [6]. Upside catalysts include continued migration to cheaper open weights, new high-performing releases, Chinese/open-model traffic growth, and enterprise concerns over sovereignty or vendor lock-in [1][2][3]. Chinese-developed models command over 70% of token share on OpenRouter by early 2026 and account for 60.5% of open-weight model use in scientific research papers [7][8].
Bearish risks persist, including a reported production gap where open models reach production 12 percentage points less often than closed models [3]. Operational, security, and compliance friction, alongside potential maintainer sustainability and relicensing pressure, pose challenges [3][9][10]. Regulatory costs are also a factor, with the 2026 CRA vulnerability-reporting requirements beginning September 11 [3]. Downside catalysts include a closed-model price/performance breakthrough, outages or safety incidents in major open models, stricter licensing/regulation, or a Vercel benchmark methodology change [1][2][3]. Despite 97% of organizations utilizing open-source software, sustainability challenges exist, such as bottlenecked patch review processes in foundational projects like the Linux kernel [11][12].

Key Dates & Catalysts

  • Expiration: October 06, 2026
  • Closes: September 29, 2026
Sources (12)
  1. 1Open Source market share on Sep 28th - Kalshi Odds | CoinRithmcoinrithm.com
  2. 2Open weight models process majority of AI tokens for the first time - TechCentral.ietechcentral.ie
  3. 3State of Open Source AI 2026stateofopensource.ai
  4. 4Open Source market share on Sep 26th - Kalshi Odds | CoinRithmcoinrithm.com
  5. 5DeepSeek market share this week - Kalshi Odds | CoinRithmcoinrithm.com
  6. 6Open Source Software Market Size, Trends Report 2026-2030thebusinessresearchcompany.com
  7. 7Who Uses Open-Weight Models? China and the Shifting Geography of AI in Sciencearxiv.org
  8. 8The ATOM Report:Measuring the Open Language Model Ecosystemarxiv.org
  9. 9Open-Source Software Report 2026: 4,496 Projects Analyzeddev.co
  10. 10What's in Store for Open Source in 2026appdevelopermagazine.com
  11. 11AlphaOpsBench: Benchmarking End-to-End Alpha Strategy Operationalization in Prediction Marketsarxiv.org
  12. 12Quantifying Competitive Relationships Among Open-Source Software Projectsarxiv.org

13. Historical Resolutions

Historical Resolutions: 20 markets in this series

Outcomes: 9 resolved YES, 11 resolved NO

Recent resolutions:

  • KXOPENSOURCESHARE-26SEP27-T90: NO (Sep 27, 2026)
  • KXOPENSOURCESHARE-26SEP27-T87: NO (Sep 27, 2026)
  • KXOPENSOURCESHARE-26SEP27-T84: NO (Sep 27, 2026)
  • KXOPENSOURCESHARE-26SEP27-T81: NO (Sep 27, 2026)
  • KXOPENSOURCESHARE-26SEP27-T78: NO (Sep 27, 2026)