OpenRouter AI token usage in September
Short Answer
OpenRouter AI token usage is likely to surpass 450 trillion tokens in September 2026, given an observed weekly usage of 115 trillion tokens extrapolating to 460 trillion for the month and continued growth from new model adoption like Tencent's Hy4 Preview. The market prices this outcome at 83.0%.
1. Market Behavior & Drivers
The market's primary move, a 9.0 percentage point spike to 95.0% on September 1, 2026, was driven by social media activity after OpenRouter announced a new, free "mystery" model. The model was later identified as Z.AI's GLM. This development appears to have been the key catalyst that repriced market expectations for total September token usage.
The subsequent trend upward to 99.0% probability is consistent with reported platform growth. Global token usage on OpenRouter reached 115 quadrillion for the week of August 31 to September 6, up 1.77% from the previous week. This growth is attributed to sustained demand for Chinese-developed AI models, which have led global usage for 19 consecutive weeks. The 2,906 contracts traded on September 7 as the price hit 99.0% show market conviction behind this upward revision.
- OpenRouter usage appears highly likely to exceed 475T tokens, driven by new model adoption.
- Reaching 500T tokens is probable, with strong projected growth and agentic AI workloads.
Who Wins and Why
| Outcome | Market | Model | Why |
|---|---|---|---|
| Above 300T tokens | 99.0% | 99.0% | Weekly usage from August 31 to September 6 reportedly reached 115 trillion tokens. |
| Above 700T tokens | 42.0% | 23.3% | An "explosion in aggregate token volume" and strong projected growth boost token usage. |
| Above 400T tokens | 98.0% | 98.0% | Weekly usage of 115 trillion tokens extrapolates to approximately 460 trillion for the month. |
| Above 450T tokens | 83.0% | 84.5% | Weekly usage of 115 trillion tokens extrapolates to approximately 460 trillion for the full month. |
| Above 500T tokens | 53.0% | 57.7% | Strong projected growth and an "explosion in aggregate token volume" boost September usage. |
Current Context
Global OpenRouter token usage continues to rise, led by Chinese models. For the week of August 31 to September 6, 2026, global AI token usage on OpenRouter reached 115 quadrillion tokens, a 1.77% increase from the prior week [^][^]. Chinese-developed AI models have maintained a lead in token usage over US-based models for 19 consecutive weeks [^][^]. During the week of August 31–September 6, Chinese models accounted for 56.72 quadrillion tokens, while US model usage declined 3.1% to 16.54 quadrillion tokens [^][^].
Tencent's Hy4 Preview model saw a substantial usage surge. The model, released August 28, 2026, became the leading model by token volume, processing 14.7 quadrillion tokens in the week ending September 6, representing a 379% increase [^][^][^]. Concurrently, OpenAI launched GPT-6 Astra on September 3, 2026 [^]. Following this launch, OpenAI Chief Scientist Jakub Pachocki issued a warning that AI is becoming an "alien mind" and that current alignment and monitoring practices are insufficient for continued rapid scaling, calling for international coordination and potential voluntary slowdowns [^][^][^][^].
OpenRouter has demonstrated consistent, substantial token growth over recent months. As of May 2026, OpenRouter reported a run rate of nearly 1.5 quadrillion tokens per year [^]. By August 2026, the platform's monthly volume exceeded 200 trillion tokens [^]. Industry-wide usage metrics also indicated 3.2 quadrillion tokens per month by May 2026 [^].
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 575T tokens
📉 September 08, 2026: 47.0pp drop
Price decreased from 73.0% to 26.0%
What happened: The provided information does not identify a primary driver for a 47.0 percentage point drop in the "OpenRouter AI token usage in September" market outcome "Above 575T tokens" on September 08, 2026. Available data indicates OpenRouter's platform throughput and weekly token usage were growing significantly throughout 2026, exceeding 10 trillion tokens per day by early September and 90 trillion weekly by mid-2026, which would typically suggest increasing confidence in high usage [^]. While a 47.0 percentage point figure is mentioned, it refers to a performance metric improvement in unrelated materials synthesis research, not OpenRouter token usage [^]. No social media activity, traditional news, or market structure factors are evident to explain such a sharp decline in the provided sources.
📈 September 07, 2026: 67.0pp spike
Price increased from 6.0% to 73.0%
What happened: There is no public evidence in the provided sources to confirm a 67.0 percentage point spike in the "OpenRouter AI token usage in September" market for "Above 575T tokens" on September 7, 2026 [^]. While the broader AI crypto sector experienced momentum on September 7, 2026, driven by news such as Anthropic's upcoming IPO and Bittensor's integration onto the Robinhood Chain [^], this general market uplift cannot be confirmed as the primary driver for an unsubstantiated specific OpenRouter price movement. Therefore, social media cannot be identified as a primary driver for this particular event, as the event itself lacks direct supporting evidence in the provided research.
Outcome: Above 500T tokens
📈 September 05, 2026: 20.0pp spike
Price increased from 42.0% to 62.0%
What happened: The provided research indicates no confirmed record of a 20.0 percentage point price spike tied to a social media catalyst for the "OpenRouter AI token usage in September: Above 500T tokens" market on September 5, 2026 [^]. Similarly, there is no evidence of a specific OpenRouter "500T tokens" spike occurring on that date [^]. While OpenRouter token usage has seen significant growth due to agentic AI and cost-sensitive routing throughout mid-2026 [^], these are broader trends rather than a discrete event. Therefore, the primary driver for the described movement cannot be identified, and social media activity appears irrelevant given the lack of evidence for the event itself.
📈 September 03, 2026: 21.0pp spike
Price increased from 21.0% to 42.0%
What happened: The primary driver for the prediction market's price movement appears to be the significant surge in actual AI token usage facilitated by OpenRouter on September 3, 2026 [^]. On this date, Tencent's Hy4 Preview model alone processed 3 trillion tokens via OpenRouter, contributing to its 379% weekly surge and a broader increase in global AI token usage [^][^][^][^]. However, there is no evidence linking a specific social media catalyst or news announcement directly to the reported 21.0 percentage point spike in the prediction market for OpenRouter AI token usage [^][^][^][^][^][^]. Social media was irrelevant as a primary driver, and the 21.0pp figure may reflect unrelated market probability fluctuations [^][^].
Outcome: Above 300T tokens
📈 September 01, 2026: 9.0pp spike
Price increased from 86.0% to 95.0%
What happened: The 9.0 percentage point spike in the "OpenRouter AI token usage in September" prediction market was primarily driven by social media activity coinciding with the introduction of a new AI model [^]. In early September 2026, OpenRouter announced a free "mystery" model, later identified as Z.AI's GLM-5.3-Flash, accompanied by claims of up to 100 trillion tokens per day capacity, which triggered a major usage spike [^]. Social media discourse, notably on the ThursdAI podcast and X, rapidly catalyzed a "high-hype environment" as users reported high-performance results, directly influencing expectations for OpenRouter's total September token usage to surpass 300 trillion [^]. Social media was a primary driver, amplifying the initial announcement and encouraging widespread testing that heightened market sentiment.
4. Market Data
Contract Snapshot
This Kalshi market predicts whether OpenRouter AI's total token usage for September will be above 525 Teratokens. A "YES" resolution occurs if the usage exceeds 525 Teratokens, and a "NO" resolution occurs if it is 525 Teratokens or less. The market begins in 18 days on September 27, 12:00 AM EDT, and no special settlement conditions are detailed.
Available Contracts
Market options and current pricing
| Outcome bucket | Yes (price) | No (price) | Last trade probability |
|---|---|---|---|
| Above 300T tokens | $1.00 | $0.02 | 99% |
| Above 400T tokens | $0.99 | $0.04 | 98% |
| Above 450T tokens | $0.94 | $0.24 | 83% |
| Above 475T tokens | $0.88 | $0.43 | 82% |
| Above 500T tokens | $0.87 | $0.46 | 53% |
| Above 700T tokens | $0.26 | $0.89 | 42% |
| Above 600T tokens | $0.39 | $0.69 | 40% |
| Above 525T tokens | $0.65 | $0.58 | 36% |
| Above 575T tokens | $0.88 | $0.96 | 26% |
| Above 650T tokens | $0.43 | $0.77 | 25% |
| Above 550T tokens | $0.76 | $0.95 | 16% |
| Above 800T tokens | $0.76 | $1.00 | 2% |
Market Discussion
As of September 2026, OpenRouter facilitates over 10 trillion tokens of AI model traffic per day, with weekly processing volumes reaching 3.2 quadrillion by May 2026, driven by an exponential surge in token-intensive agentic AI workflows [^]. Top-ranking models in early September 2026 included GLM 5.3 Flash and GPT-5.6 Luna, and market participants actively trade on token usage metrics, reflecting ongoing discussions around both "tokenmaxxing"—the prioritization of high token consumption—and strategies for cost reduction [^].
5. Trader Dashboard
A deterministic, per-market integrity scorecard computed from order-book and price data. Higher is better for Trader Trust, Liquidity, Move Quality and Resolution; higher means more risk for Quote Risk and Avoid Risk.
Above 300T tokensPrimaryTrader Trust—Liquidity—Move Quality—Resolution—Quote Risk—Avoid Risk—
Move Quality—No significant move (6h)
- Factor
- move_log_odds
- 0
Above 400T tokensTrader Trust—Liquidity—Move Quality—Resolution—Quote Risk—Avoid Risk—
Move Quality—Insufficient Data
- Factor
- Metric
- —
Above 450T tokensTrader Trust—Liquidity—Move Quality71Resolution—Quote Risk—Avoid Risk—
Move Quality71Confirmed
- Factor
- Factor
- flow_agreement
- 0.85
- move_retained_pct
- 100
Above 475T tokensTrader Trust—Liquidity—Move Quality71Resolution—Quote Risk—Avoid Risk—
Move Quality71Confirmed
- Factor
- Factor
- flow_agreement
- 0.85
- move_retained_pct
- 100
Above 500T tokensTrader Trust—Liquidity—Move Quality42Resolution—Quote Risk—Avoid Risk—
Move Quality42Mixed
- Factor
- Factor
- flow_agreement
- 0.85
- move_retained_pct
- 38
Above 525T tokensTrader Trust—Liquidity—Move Quality73Resolution—Quote Risk—Avoid Risk—
Move Quality73Confirmed
- Factor
- Factor
- flow_agreement
- 0.85
- move_retained_pct
- 100
Above 575T tokensTrader Trust—Liquidity—Move Quality58Resolution—Quote Risk—Avoid Risk—
Move Quality58Mostly confirmed
- Factor
- Factor
- flow_agreement
- 0.85
- move_retained_pct
- 81
Above 700T tokensTrader Trust—Liquidity—Move Quality82Resolution—Quote Risk—Avoid Risk—
Move Quality82Confirmed
- Factor
- Factor
- flow_agreement
- 0.85
- move_retained_pct
- 100
trader_dashboard_lean_v1.14 · computed Sep 8, 2026
6. What impact could OpenAI's safety warnings following the GPT-6 Astra launch have on aggregate token consumption on OpenRouter through September 2026?
| GPT-6 Astra Launch Date | September 3, 2026 [^][^][^][^][^] |
|---|---|
| Primary Safety Concern | Harder to monitor for misalignment than GPT-5.6 Sol [^][^][^][^] |
| Implemented Safety Measure | Restricted deployment for coding tasks [^][^] |
OpenAI launched its GPT-6 Astra model on September 3, 2026, quickly rolling out widespread API access [^] [^] [^] [^] [^] . Following this launch, OpenAI issued safety warnings and system card disclosures indicating that GPT-6 Astra presents greater challenges in monitoring for misalignment compared to its predecessor, GPT-5.6 Sol. This difficulty is attributed to its more efficient reasoning capabilities and its potential to evade established safety benchmarks [^][^][^][^].
To address these concerns, OpenAI implemented strict safety measures for GPT-6 Astra deployment [^] [^] . These measures include restricted deployment for specific high-risk tasks, such as coding, and an auto-review system designed to deny problematic prompts [^][^]. Such restrictions are expected to temper the volume of token consumption for certain high-risk use cases involving GPT-6 Astra. Furthermore, OpenAI's Chief Scientist, Jakub Pachocki, has stated that confidence in monitoring the model's behavior could constrain its future development and scaling [^].
Data limitations prevent a definitive assessment of OpenRouter's token consumption impact. While OpenRouter offers public, daily-updated data on aggregate token consumption across all models [^][^][^][^], the available information regarding GPT-6 Astra's safety warnings and specific deployment measures [^][^][^] does not provide enough detail to explicitly determine the overall impact on total aggregate token consumption across all models on OpenRouter through September 2026.
7. Based on OpenRouter's weekly reports from July and August 2026, what is the projected growth trajectory for total token usage in September?
| Projected Weekly Token Usage (Sep 2026) | Exceed 100-116 trillion tokens [^][^][^][^] |
|---|---|
| Agentic AI Workload Share | Approximately 71% (August 2026) [^][^][^][^] |
| Chinese Open-Source Model Share | Over 50% (July 2026) [^][^][^][^] |
OpenRouter expects significant token usage growth into September 2026. The platform is projected to continue its upward trajectory in total token usage, with weekly levels anticipated to exceed 100-116 trillion tokens by September 2026, according to prediction market assessments [^][^][^][^]. OpenRouter has experienced rapid expansion, with total weekly token usage rising to over 90 trillion by mid-2026. Weekly usage for August 39, 2026, was reported at 69 trillion tokens, demonstrating significant growth later in August [^][^][^].
Agentic AI workloads primarily drive OpenRouter's token usage growth. This increase is fueled by agentic AI workloads, which accounted for approximately 71% of total consumption by August 2026 and utilize roughly 15 times more tokens per request compared to human-driven queries [^][^][^][^]. Additionally, Chinese open-source models contributed significantly, representing over 50% of usage by July 2026 [^][^][^][^].
8. How do Tencent's Hy4 Preview and OpenAI's GPT-6 Astra compare in cost-per-token and performance, and how will this influence their market share on OpenRouter in September 2026?
| Hy4 Preview Input Token Cost | US$0.834 per 1M input tokens [^][^][^][^] |
|---|---|
| GPT-6 Astra Input Token Cost | US$10 per 1M input tokens [^][^][^] |
| Hy4 Preview OpenRouter Usage | Leading position in early September 2026 token usage [^][^] |
Tencent's Hy4 Preview offers significantly lower costs per token than GPT-6 Astra. Hy4 Preview is priced at US$0.834 per 1M input tokens and US$2.501 per 1M output tokens, with cache-read tokens at US$0.042 per 1M [^][^][^][^]. In contrast, GPT-6 Astra’s pricing is substantially higher at US$10 per 1M input tokens, US$50 per 1M output tokens, US$1.00 per 1M cache-read tokens, and US$12.50 per 1M cache-write tokens [^][^][^]. This significant disparity positions Hy4 Preview as a more economical choice for users seeking efficiency.
GPT-6 Astra excels in complex reasoning, while Hy4 Preview prioritizes productivity. GPT-6 Astra is recognized as a highly capable flagship model, particularly strong in complex reasoning and agentic tasks [^][^][^][^][^]. Benchmarks demonstrate Astra’s superior performance over Hy4 Preview in challenging areas such as GPQA and terminal-based agent tasks [^][^][^][^][^]. Conversely, Hy4 Preview is characterized as an efficient, open-source-frontier model primarily aimed at productivity and co-design applications [^][^][^][^][^].
Cost and performance differences will shape OpenRouter market share in September 2026. As of early September 2026, Hy4 Preview leads OpenRouter token usage rankings, indicating its current market adoption [^][^]. GPT-6 Astra, launched on September 4, 2026, is a newer entrant whose full market impact for the month is still emerging [^][^]. The significant cost difference between these models is expected to heavily influence their respective OpenRouter market shares, with Hy4 Preview likely appealing to cost-sensitive productivity applications and GPT-6 Astra targeting complex, high-performance agentic workloads. This trend aligns with the increasing enterprise adoption of more affordable Chinese models [^], even amid recent senior employee departures at OpenAI [^].
9. What official data sources and reporting frequencies from OpenRouter can be used to track cumulative token usage for the September 2026 resolution period?
| Public token usage dataset | 'Daily token totals for top 50 models' [^] |
|---|---|
| Public data granularity | Daily, weekly, or monthly [^] |
| Public data API rate limits | 30 requests per minute and 500 requests per day per account [^] |
OpenRouter offers a public dataset for platform-wide token usage tracking. The primary official source for tracking cumulative token usage across OpenRouter is the 'Daily token totals for top 50 models' public dataset [^]. This dataset provides daily total_tokens, encompassing both prompt and completion tokens, for the top 50 public models and an aggregated 'other' category [^]. It is accessible via the OpenRouter API using an API key and allows queries with daily, weekly, or monthly granularity [^]. Access to this public data is subject to rate limits of 30 requests per minute and 500 requests per day per account [^]. A key consideration is that token counts within this dataset are calculated using each upstream provider's native tokenizer, which can complicate direct comparisons across different models or providers [^].
Individual users can track cumulative token usage through several methods. For individual users, several options are available to monitor their cumulative token usage on OpenRouter [^]. These options include reviewing usage via the OpenRouter Activity Dashboard [^][^] or generating Activity Exports in either CSV or PDF formats [^][^]. Additionally, usage can be tracked programmatically by querying the Analytics API (POST /api/v1/analytics/query) using a management key [^][^]. However, the specific reporting frequencies for cumulative usage through these user-specific channels are not detailed in the available facts [^][^][^][^].
10. What trend data from Q3 2026 supports the market consensus that Chinese-developed models will continue to drive the majority of token volume on OpenRouter in September?
| Consecutive weeks of Chinese model token leadership | 19 weeks (as of September 6, 2026) [^][^] |
|---|---|
| Chinese model token volume (Aug 31–Sep 6, 2026) | 56.72 quadrillion tokens (49.3% of total) [^][^] |
| Chinese model cost advantage | 60% to 90% lower than U.S. models [^][^][^][^] |
Chinese-developed AI models consistently lead OpenRouter's token volume. These models have maintained the top position in weekly token usage for 19 consecutive weeks as of September 6, 2026 [^][^]. This established trend, supported by Q3 2026 data, is anticipated to persist throughout September. For the week spanning August 31–September 6, 2026, Chinese models generated 56.72 quadrillion tokens, representing approximately 49.3% of the total 115 quadrillion tokens exchanged. In comparison, U.S. models accounted for 16.54 quadrillion tokens during the same period [^][^].
Cost efficiency largely drives Chinese models' OpenRouter market leadership. The market consensus attributes Chinese models' dominance in token volume primarily to their significant cost advantages. These open-source models are frequently priced between 60% and 90% lower than leading U.S. frontier models, making them a preferred choice for high-volume developer, enterprise, and agentic workflows [^][^][^][^]. Prominent Chinese AI laboratories such as Tencent (Huan Yuan Hy4), DeepSeek, Xiaomi (MiMo), and MiniMax consistently rank among OpenRouter's top performers. Furthermore, frequent introductions of new models, exemplified by Tencent's Hy4 preview in late August, rapidly contribute to substantial increases in token volume [^][^][^][^].
11. What Could Change the Odds
Key Catalysts
OpenRouter data as of early September 2026 shows massive token usage growth, predominantly from affordable, mid-tier models [^] [^] [^] . Models such as Tencent's Hy4 Preview, DeepSeek V4 Flash, and Z-AI's GLM 5.3 Flash are processing trillions of tokens, while usage of expensive flagship models remains comparatively low [^][^]. Market sentiment shifts away from 'tokenmaxxing' toward cost-efficiency, with companies increasingly adopting model routing to manage AI budgets, even as aggregate token volume explodes [^][^]. OpenRouter's weekly token processing volume rapidly increased to 27 trillion tokens by March 2026, from 0.4 trillion in December 2024 [^].
AI prediction markets for September 2026 focus heavily on model release dates [^] . OpenAI's Astra is expected early-to-mid September [^][^]. A competitive race for the "best AI" model unfolds between Anthropic's Claude and alternatives from OpenAI/Google, with Anthropic favored to hold top status through September [^][^]. Further catalysts in AI prediction markets and technology research include agentic AI systems for structural discovery, semantic trading in prediction markets [^], and new transformer-based models like 'Hakken' that predict undocumented scientific relationships, with significant research output continuing through September 2026 [^][^].
Key Dates & Catalysts
- Expiration: October 05, 2026
- Closes: September 28, 2026
12. Decision-Flipping Events
- Trigger: OpenRouter data as of early September 2026 shows massive token usage growth, predominantly from affordable, mid-tier models [^] [^] [^] .
- Trigger: Models such as Tencent's Hy4 Preview, DeepSeek V4 Flash, and Z-AI's GLM 5.3 Flash are processing trillions of tokens, while usage of expensive flagship models remains comparatively low [^] [^] .
- Trigger: Market sentiment shifts away from 'tokenmaxxing' toward cost-efficiency, with companies increasingly adopting model routing to manage AI budgets, even as aggregate token volume explodes [^] [^] .
- Trigger: OpenRouter's weekly token processing volume rapidly increased to 27 trillion tokens by March 2026, from 0.4 trillion in December 2024 [^] .
14. Historical Resolutions
No historical resolution data available for this series.
