forecasting
agent
← back to scans

Will OpenRouter AI token usage for October 2026 (Sep 28–Oct 25) be above 550T tokens?

kalshiScience and Technology·2026-10-02·view on kalshi ↗
70%
Agent
46%
Market Price
+23.8%
Edge
56%
Confidence
2,562
Volume
72.0c
Spread
23
Days to resolution
12
Markets in event
final rationale
Mean of 5 draws (0.72, 0.72, 0.77, 0.64, 0.64) after one crux research pass. (0) Not met yet: window Sep 28–Oct 25 just started; OpenRouter rankings only through Sep 30–Oct 1 with no window aggregate. (a) ~23 days remain; resolution observed 10 AM ET Oct 26. (b) Status quo: 550T/28 days needs ~19.6T/day (~137.5T/week). Last confirmed week (Sep 21–27) was ~146T, after 126.8T and 128.9T; the week through Sep 30 shows top-10 models alone at ~107T, implying ~150–165T/week with the long tail. Flat run-rate gives ~580–640T, a ~5–15% cushion. (c) NO: stealth model Space Bunny Alpha (28.4T/week, ~5T/day) is withdrawn or ends free access, and launch-spike models keep fading (GLM -44%, Hy4 -42% WoW, DeepSeek share -2%), dragging the weekly average under ~137T; plus resolution-source aggregation risk. (d) YES: trend growth continues (series ~4x May→Sep), DeepSeek V4.1 Flash keeps growing, Oct 1 launches add volume, landing ~600T+. Market anchor: live 46% on a 27/99 spread with a 97% last close is thin and noisy (mid ~63%), so I weight fundamentals more, discounted for stealth-model and spike-reversal risk.
price vs. prediction
market price agent forecast at t+0
16 follow-ups
1. Planner Opus 5.5 · picks the research tools
Tools chosen: kalshi_direct resolution_source kalshi_related polymarket_related
Planner reasoning
The market resolves on OpenRouter's AI Model Rankings page, so resolution_source is the key data point for recent weekly token volumes and the trend. Kalshi_direct gives the price anchor, and kalshi_related covers the sibling strike markets and prior months' results, which show how fast usage is growing. Polymarket_related checks whether the other venue has an independent price on the same question.
2. Research parallel tool calls + 4 web research loops (Sonnet 5) · read raw by the forecasters
at a glance
RESEARCH DIGEST (built in code: tool summaries and each report's bottom line; the forecasters read the full tool outputs and reports) Tools - kalshi_direct: ## This Market's Kalshi Data **Above 550T tokens** (KXTOKENUSEM-26OCT26-T550) - Current price (probability): 46.00% (live; bid 27% / ask 99%) - Last daily close (2026-10-02): 97.00% - 7-day price change: -51.00% - 30-day price change: -51.00% - Average daily volume: 2541 contracts - Price range: 97.00% - 97.00% - Data points: 1 days - resolution_source: already_resolved=unclear; matches_criteria=True; last_updated=Oct 1, 2026. The page shows usage data through Oct 1, 2026, with a 'This Month' trailing-30-day leaderboard listing only the top 10 models (e.g., GLM 5.3 Flash 56.5T, Hy4 preview 55.1T, DeepSeek V4.1 Flash 54.9T, etc.), not a total token usage figure across all models, and this period does not match the Oct 2026 (Sep 28–Oct 25) window d - kalshi_related: 0 related markets / summaries. series KXTOKENUSEM: 0 markets (skipped 12 no-signal) | keyword 'OpenRouter token usage': no matches - polymarket_related: Polymarket search over 1 keyword(s), kept 5 matches. keyword 'OpenRouter': 5 markets Research loops, bottom lines - current state: The most recent confirmed weekly OpenRouter run-rate (~146T tokens for the week of Sep 21, 2026, up from ~127-129T in the prior two weeks) implies a 4-week total for Sep 28–Oct 25 in the roughly 540–630T range if the recent growth trajectory holds, putting the 550T threshold right at/above the current run-rate's trend line — a close call that leans toward Yes if the acceleration seen in the Sep 21 week continues, but could fall short if the jump to 145.8T was a one-off spike rather than a sustai - base rate: The only comparable prior instance of this exact contract type (the September 2026 OpenRouter monthly-token Kalshi market) saw the market price the "Above 500T" strike at roughly 53-58% with weeks left to run, while usage continued compounding rapidly (volumes roughly quadrupled from May to September 2026); given October's run-rate entering the window is higher still (weekly leader already 28T+ vs ~12T a month prior), the base-rate/outside-view evidence leans toward October 2026 usage likely cle - yes case: The freshest primary-source data (OpenRouter's own rankings fetched Oct 2, plus three independent weekly-tracker rebuilds through Sep 30) show a sustained, multi-vendor acceleration — weekly tokens already at 145.8T (up 47% y/y-equivalent) before the window opened, a single day's top-10 models alone totaling ~16.7T tokens, and fresh stealth/model launches landing exactly at the start of October — all of which point toward a run-rate comfortably capable of clearing 550T over Sep 28–Oct 25, especi - no case: The freshest data point available (DeepSeek's first negative weekly growth reading, reported Oct 1) plus a stalled weekly-tracker figure unchanged across two consecutive builds spanning the market's open, several top-10 models already shrinking month-over-month, and a 28-day resolution window that is structurally shorter than the 30-day "This Month" metric most extrapolations rely on, together support a case that October's total lands at or below the 550T line rather than comfortably above it. Crux follow-up - The freshest available data (through Sep 30) shows mixed but on-net still-elevated usage — individual spike-driven models like GLM and Hy4 are fading as the "NO" case anticipated, but this is offset by continued strong growth in Space Bunny Alpha and DeepSeek V4.1 Flash, with top-10 daily volume (~16.7T/day) still tracking near or above the ~19.6T/day breakeven pace once the long tail is added. This modestly favors the "YES" (above 550T) side of the fundamentals debate, though not decisively, si
tools
toolstatustimeitemssummary
kalshi_direct OK 0.2s 1 ## This Market's Kalshi Data **Above 550T tokens** (KXTOKENUSEM-26OCT26-T550) - Current price (probability): 46.00% (live; bid 27% / ask 99%) - Last daily close (2026-10-02): 97.00% - 7-day price change: -51.00% - 30-day price change: -51.00% - Average daily volume: 2541 contracts - Price range: 97
resolution_source OK 6.3s 1 already_resolved=unclear; matches_criteria=True; last_updated=Oct 1, 2026. The page shows usage data through Oct 1, 2026, with a 'This Month' trailing-30-day leaderboard listing only the top 10 models (e.g., GLM 5.3 Flash 56.5T, Hy4 preview 55.1T, DeepSeek V4.1 Flash 54.9T, etc.), not a total token
kalshi_related OK 0.5s 0 0 related markets / summaries. series KXTOKENUSEM: 0 markets (skipped 12 no-signal) | keyword 'OpenRouter token usage': no matches
polymarket_related OK 0.4s 5 Polymarket search over 1 keyword(s), kept 5 matches. keyword 'OpenRouter': 5 markets
research loops · 4/4 ok · 17 searches · 7 page reads
dimensionstatustimesearchespage readsreport
current state OK 55s 5 3 4798 chars
base rate OK 37s 4 2 5189 chars
yes case OK 43s 5 1 6015 chars
no case OK 53s 3 1 5297 chars
report: current state
- [2026-10-02] OpenRouter's official rankings page shows usage data aggregated into daily UTC buckets, with the "This Month" view covering trailing 30 days , and as of this fetch the page stated "Usage data through Sep 30, 2026." (https://openrouter.ai/rankings) - [2026-10-02] Per OpenRouter's own trailing-30-day ("This Month") leaderboard (period ending Sep 30, 2026), the top 10 models alone totaled roughly 359T tokens, led by GLM 5.3 Flash (56.9T, +600%) and Tencent Hy4 preview (56.1T, >999%), with DeepSeek V4.1 Flash (51.4T, new) and GPT-5.6 Luna (51.3T, +111%) close behind (https://openrouter.ai/rankings). - [2026-09-30] An independent weekly tracker (token-elasticity dashboard, built Sep 30) reported weekly tokens (OpenRouter) of 145.8T for the data week of 2026-09-21 , an acceleration from 128.9T for the week of 2026-09-14 and 126.8T for the week of 2026-09-07 (https://github.com/mbhardwaj103/token-elasticity/issues/17). - [2026-09-17] the-decoder.com reported that weekly token consumption on OpenRouter had surged more than 25,000 percent since January 2025, from 0.5 trillion to 126.2 trillion tokens , underscoring the accelerating, reasoning/agent-driven growth curve feeding into the October window (https://the-decoder.com/openrouters-staggering-token-chart-is-the-ai-bubble-debate-in-a-single-image/). - [2026-09-30] A third-party rankings mirror (tokenmaxxing.com), sourced from OpenRouter's public API with data through Sep 30, 2026, shows the top single model (Space Bunny Alpha) processing 5.3T tokens on 2026-09-30 alone, with a 28.7T 30-day total , consistent with OpenRouter's own figures (https://tokenmaxxing.com/openrouter-rankings). - [2026-10-01] OpenRouter's models page shows continued new model launches dated Oct 1, 2026 (e.g., a Liquid model and an Unbiased "Pareto 26.10 Preview"), indicating fresh model releases that historically drive usage spikes are landing right at the start of the October measurement window (https://openrouter.ai/models). - [2026-09-08] A prior Kalshi-linked research report (Octagon AI) on the September 2026 token-usage market noted that for the week of Aug 31–Sep 6, 2026 global AI token usage on OpenRouter reached 115 quadrillion [trillion] tokens, a 1.77% increase from the prior week , and that market odds for "Above 500T tokens" (for September) stood near 53-58% at that time (https://www.octagonai.co/markets/science-and-technology/tokens/openrouter-ai-token-usage-in-september). - [2026-09-30] Market-share data on OpenRouter's rankings page (week beginning Sep 21, 2026) shows "Share of text requests made on OpenRouter in the week beginning Sep 21, 2026" with DeepSeek leading at 24.4% of requests, Google at 20.5% (+12% WoW) and OpenAI at 17.9% (+6% WoW) — signaling broad-based, multi-vendor growth rather than a single model driving the trend (https://openrouter.ai/rankings). - [2026-09-17] the-decoder also noted that "OpenAI's GPT 5.6 Luna recently dominated token consumption on OpenRouter, though that doesn't necessarily mean more people are using it... The model may simply generate more tokens per prompt" , a reminder that token totals are inflated by verbose reasoning/agentic models, a structural factor pushing monthly totals higher independent of "real" usage growth (https://the-decoder.com/openrouters-staggering-token-chart-is-the-ai-bubble-debate-in-a-single-image/). - [2026-10-01] No week-of-Sep-28 data point has yet appeared in the token-elasticity tracker as of the Oct 1 build (still showing "Data week of 2026-09-21"), meaning the most recent confirmed weekly figure covering the start of the market window itself is not yet published; based on the historical ~7-9 day reporting lag, a Sep28 week figure should surface around Oct 5-7 (https://github.com/mbhardwaj103/token-elasticity/issues/17). BOTTOM LINE: The most recent confirmed weekly OpenRouter run-rate (~146T tokens for the week of Sep 21, 2026, up from ~127-129T in the prior two weeks) implies a 4-week total for Sep 28–Oct 25 in the roughly 540–630T range if the recent growth trajectory holds, putting the 550T threshold right at/above the current run-rate's trend line — a close call that leans toward Yes if the acceleration seen in the Sep 21 week continues, but could fall short if the jump to 145.8T was a one-off spike rather than a sustained new baseline. NOT FOUND: No official OpenRouter-published single aggregate "total platform tokens for October 2026" figure exists yet (the window is still in progress); no confirmed weekly figure for the week of Sep 28, 2026 was available as of this research; could not confirm whether the 115 "quadrillion" figure cited by Octagon AI was a labeling error for trillion (it appears inconsistent with other trillion-based weekly figures from the same period).
report: base rate
- [2026-08-26] Platform run-rate entering Q3 2026 was already in the ~90-115T tokens/week range (news trackers put the week of Aug 31–Sep 6 at Global token usage on OpenRouter reached 115 quadrillion for the week of August 31 to September 6, up 1.77% from the previous week , cryptobriefing.com (https://cryptobriefing.com/openrouter-9000x-token-usage-growth/). - [2026-05-26] OpenRouter's own disclosed scale a few months earlier was lower: OpenRouter said its traffic has climbed to 25 trillion tokens per week, or roughly 100 trillion tokens per month, marking a 5x increase from six months ago , techstartups.com (https://techstartups.com/2026/05/26/openrouter-raises-113m-as-ai-token-usage-surges-to-100-trillion-monthly/). This shows the series roughly quadrupled from May (~100T/month) to early September (~460-500T/month), i.e. explosive, non-linear monthly growth over the preceding quarter. - [2026-09-08] For the sibling September 2026 Kalshi contract (same series, "Above 500T tokens"), the Octagon market tracker showed the market/model pricing Above 500T tokens | 53.0% | 57.7% | Strong projected growth and an "explosion in aggregate token volume" boost September usage. as of Sep 8, implying forecasters saw September landing close to, but not certainly above, 500T — i.e., a coin-flip-ish outcome for the 500T line a few weeks before close, source octagonai.co (https://www.octagonai.co/markets/science-and-technology/tokens/openrouter-ai-token-usage-in-september). - [2026-09-08] The same tracker shows the market for "Above 575T tokens" swung wildly (73%→26% drop noted, with no clear underlying driver identified), illustrating that higher-threshold outcomes in this series have historically been volatile and thinly traded rather than settling smoothly — a caution for extrapolating precise base rates from market prices alone, octagonai.co (https://www.octagonai.co/markets/science-and-technology/tokens/openrouter-ai-token-usage-in-september). - [2026-10-01] A third-party tracker computing OpenRouter's own "This Month" (trailing-30-day) leaderboard as of Sep 30, 2026 lists top named models' 30-day totals (GLM 5.3 Flash 56.9T, Hy4 preview 56.1T, DeepSeek V4.1 Flash 51.4T, GPT-5.6 Luna 51.3T, DeepSeek V4 Flash 0731 42.5T, Space Bunny Alpha 28.7T, MiMo-V2.5 19.9T, Nemotron 3 Ultra 19.2T, DeepSeek V4 Flash 0423 17.3T, Hy3 15.6T) — summing to roughly 359T tokens just among the top-10 named models, before counting the long tail of hundreds of other smaller models that make up the "Others" category on OpenRouter's own market-share chart, tokenmaxxing.com (https://tokenmaxxing.com/leaderboard). - [2026-10-01] OpenRouter's own live rankings page (usage data through Sep 30, 2026) confirms Favicon for stealth Space Bunny Alpha by stealth 28.4T tokens >999% as the single largest model by weekly tokens, with the top 10 weekly models summing to roughly 107T tokens for that one week alone, which — extrapolated across a 4-week window — is consistent with monthly totals in the 400-550T+ range even before counting the long tail, openrouter.ai/rankings (https://openrouter.ai/rankings). - [2026-09-02] An independent snapshot taken a month earlier (usage through Sep 1, 2026) found the then-current weekly leader at only DeepSeek V4 Flash 0731 led with 12.1 trillion tokens. GLM 5.3 Flash followed on 10 trillion, and OpenAI's GPT-5.6 Luna was third on 9.52 trillion — roughly a quarter of the per-model weekly volumes seen a month later, confirming the extremely steep month-over-month acceleration in total usage, webmoves.net (https://www.webmoves.net/openrouter-ai-model-rankings-real-world-usage/). - [2026-08-04] Earlier in the series' history (for the September contract's own run-up), the Kalshi/Octagon forecaster's base case was that the market would clear 450T and probably 500T tokens for a single calendar month, but was still well short of certainty on 575T+ thresholds, reflecting a "very likely to clear the 400-500T line, genuinely uncertain above 550-575T" pattern for one-month-ahead OpenRouter volume contracts, octagonai.co (https://www.octagonai.co/markets/science-and-technology/tokens/openrouter-ai-token-usage-in-september). BOTTOM LINE: The only comparable prior instance of this exact contract type (the September 2026 OpenRouter monthly-token Kalshi market) saw the market price the "Above 500T" strike at roughly 53-58% with weeks left to run, while usage continued compounding rapidly (volumes roughly quadrupled from May to September 2026); given October's run-rate entering the window is higher still (weekly leader already 28T+ vs ~12T a month prior), the base-rate/outside-view evidence leans toward October 2026 usage likely clearing 550T, but with real historical volatility in how these specific monthly-token contracts have resolved at higher thresholds. NOT FOUND: No official OpenRouter-published single-figure "total platform tokens for September 2026" (the actual resolved total for the sibling contract) could be located, nor an explicit historical settlement value/resolution record for the September 2026 Kalshi "KXTOKENUSEM" contract confirming whether it settled Yes or No at the 500T or 550T lines.
report: yes case
YES means: OpenRouter's own AI Model Rankings show total platform token usage for the Sep 28–Oct 25, 2026 UTC window exceeding 550 trillion tokens, as observed at 10:00 AM ET on Oct 26, 2026. - [2026-10-02] On OpenRouter's live rankings page (data through Sep 30, 2026), the "Today" tab's top-10 models alone processed tokens processed through the OpenRouter API by millions of developers summing to roughly 16.7T tokens in a single day — led by stealth model Space Bunny Alpha at 5.33T (+2%) and DeepSeek V4.1 Flash at 3.49T (-11%), implying a daily platform run-rate (before the long "Others" tail) alone capable of clearing ~470T over the 28-day window (https://openrouter.ai/rankings). - [2026-10-02] The same live page's "This Month" (trailing-30-day) tab shows GLM 5.3 Flash up Share of text requests made on OpenRouter in the week beginning Sep 21, 2026 with DeepSeek leading at 24.4% of requests and Google's share rising +12% week-over-week, confirming usage growth is broadening across vendors rather than concentrated in one fading model (https://openrouter.ai/rankings). - [2026-10-01] A GitHub-based independent tracker's three separate rebuilds between Sep 28–30 (issues #15, #16, #17) each confirm the week of Sep 21 held at 145.8T tokens, with token growth vs. four weeks earlier holding at 47.4% — a sustained, not one-off, acceleration carrying directly into the question's measurement window (https://github.com/mbhardwaj103/token-elasticity/issues/17). - [2026-08-19] Stripe officially agreed to acquire OpenRouter, with Stripe noting it has been helping companies optimize their token costs and route tokens efficiently, launching products like Token Billing , a strategic tie-up that could accelerate enterprise token routing onto OpenRouter's platform just as the October window opens (https://stripe.com/newsroom/news/stripe-agrees-to-acquire-openrouter). - [2026-08-19] CNBC reported the deal size at roughly $7.5 billion, with Stripe CEO Patrick Collison stating the companies would maximize profitability by routing their requests intelligently and spending their tokens efficiently — new capital and platform integration that supports continued growth in routed volume (https://www.cnbc.com/2026/08/19/stripe-openrouter-fintech-ai-model-marketplace-.html). - [2026-08-24] A precedent for sudden volume spikes: runtimewire.com reported that stealth coding model Ox Alpha generated 11.6 trillion tokens during its first three full days on that platform, making it the largest model launch in OpenRouter's history , illustrating how a single new stealth release can add tens of trillions of tokens within days — directly relevant since Space Bunny Alpha (a similar stealth model) is already OpenRouter's top daily model entering October (https://runtimewire.com/article/anonymous-ox-alpha-processes-26t-tokens-on-opencode-breaks-openrouter-launch-rec). - [2026-10-01] New model launches are landing right at the start of the measurement window: Unbiased's "Pareto 26.10 Preview" went live Oct 1, 2026 with a four-times-larger context window, a lower price on the routed catalogue , a change historically associated with higher per-request token consumption (https://orcarouter.ai/blog/unbiased-pareto-26-10-preview). - [2026-10-01] The Neuron's daily AI digest confirms at least two more new OpenRouter model listings went live Oct 1, 2026 (Inception's Mercury Decide and Liquid's "D1" decision model), adding fresh model inventory right at the window's start that can quickly capture token share as happened with prior stealth launches (https://www.theneuron.ai/digest/everything-that-happened-in-ai-today-thursday-october-1-2026/). - [2026-09-23] A third-party rankings summary (teamday.ai) using OpenRouter's own weekly data for Sep 17–22 found the top two models (GLM 5.3 Flash and DeepSeek V4.1 Flash) each processed about 18 trillion tokens in the latest weekly rankings , confirming independently that per-model weekly volumes had already reached levels consistent with a >500T/month aggregate run-rate before the October window even began (https://www.teamday.ai/blog/top-ai-models-openrouter-2026). - [2026-09-18] The same token-elasticity tracker's earliest September builds (issues #1–#6) show the week of Sep 7 at 126.8T tokens with 73.1% token growth vs. four weeks earlier, establishing a clear, multi-week-confirmed upward trend (126.8T → 128.9T → 145.8T across successive weeks) feeding directly into the start of the October window (https://github.com/mbhardwaj103/token-elasticity/issues/6). - [2026-10-02] OpenRouter's rankings methodology confirms the "This Month" figure is a trailing 30-day window ending on the most recent complete daily bucket, so the official October 2026 contract figure will capture the full momentum of the Sep-Oct ramp described above Today, This Week, and This Month each cover a trailing window of one day, seven days, and thirty days, ending with the most recent complete daily bucket (https://openrouter.ai/rankings). BOTTOM LINE: The freshest primary-source data (OpenRouter's own rankings fetched Oct 2, plus three independent weekly-tracker rebuilds through Sep 30) show a sustained, multi-vendor acceleration — weekly tokens already at 145.8T (up 47% y/y-equivalent) before the window opened, a single day's top-10 models alone totaling ~16.7T tokens, and fresh stealth/model launches landing exactly at the start of October — all of which point toward a run-rate comfortably capable of clearing 550T over Sep 28–Oct 25, especially if even one more stealth-model launch repeats the Ox Alpha-style 11T+ multi-day surge seen in August. NOT FOUND: No official OpenRouter-published single aggregate "total tokens for October 2026" figure exists yet (window still open); no confirmed weekly tracker figure for the week of Sep 28, 2026 had posted as of this research (expected ~Oct 5-7); could not confirm the precise "Others" long-tail total that OpenRouter's Market Share/Top Models chart groups separately from named models.
report: no case
YES means: OpenRouter's published AI Model Rankings token-usage total for the Sep 28–Oct 25, 2026 window, as observed at 10:00 AM ET on Oct 26, 2026, is above 550 trillion tokens. - [2026-10-01] The most watched leading indicator for this exact window — DeepSeek's weekly request share — showed DeepSeek with a strong 24.4% share of requests but also a 2% week-over-week decline for the week of September 21, which traders interpreted as a sign of potential deceleration , the first negative growth signal in the run-up to the October window (octagonai.co). - [2026-10-01] That same deceleration was broad enough to spook pricing on fast-growth strikes: contracts like "Above 23.7%" had reached 94% before being confronted with the new data showing a reversal in request volume , evidence that the recent acceleration narrative is not uniformly holding across the leaderboard (octagonai.co). - [2026-09-28/29] An independent weekly tracker's two most recent builds, one hour apart in dating (Sep 28 and Sep 29), both still show the identical figure of Weekly tokens (OpenRouter) 145.8T unchanged , meaning no new incremental weekly data point confirming continued acceleration has actually posted yet for the week straddling the market's Sep 28 open (github.com/mbhardwaj103/token-elasticity). - [2026-10-01] OpenRouter's own trailing-30-day leaderboard (through Sep 30) shows several previously dominant models now shrinking, not growing: Hy3 is down "-55%", DeepSeek V4 Flash 0423 is down "-24%", MiMo-V2.5 is down "-32%", and DeepSeek V4 Flash 0731 is down "-8%" month-over-month — four of the top ten named models are in outright decline (openrouter.ai/rankings), meaning October's total depends heavily on only two or three new entrants (Hy4 preview, DeepSeek V4.1 Flash, Space Bunny Alpha) sustaining their launch-week spikes rather than fading like their predecessors did. - [2026-10-01] A third-party mirror sourced directly from OpenRouter's API independently confirms this decomposition, describing the leaderboard explicitly as a trailing "30-day" measure, i.e. "Token usage from OpenRouter's live model rankings, through 2026-09-30... Daily tokens are the source-day total; the 30-day column is a real measured total across the trailing 30 days" (tokenmaxxing.com) — a 30-day window, not the market's actual 28-day Sep28–Oct25 span, so any extrapolation from "This Month" totals already overstates the true resolution-period total by roughly 2 days' worth of tokens (plausibly 8-10T+ at current run-rates). - [2026-09-08, re-confirmed 2026-10-01] The only prior instance of this contract type shows strike-labeling inconsistency and volatility that undermines confidence in clean resolution at a 550T-style high threshold: one Octagon snapshot frames the September contract as "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" while the same tracker elsewhere discusses a 500T strike, and notes the market for "Above 575T tokens" on September 08, 2026 saw a 47-point drop with "no social media activity, traditional news, or market structure factors" identified to explain it — i.e., high-threshold strikes in this exact series have historically repriced on noise rather than clean fundamentals, and even the September 500T strike (one rung below October's 550T) was only priced at 83% on Sep 8, not a lock (octagonai.co). - [2026-09-30] OpenRouter's own market-share chart for the week beginning Sep 21 cited by colleagues shows growth was already decelerating for the dominant vendor even as smaller players gained share — consistent with a maturing, more fragmented usage base rather than uniform platform-wide acceleration, which raises the odds that the long tail "Others" category (needed to push the aggregate over 550T on top of the ~359T from named top-10 models) underperforms a naive extrapolation. - [2026-10-01] No official OpenRouter-published single aggregate figure for any full historical month has been located by any team member to date, meaning this market's resolution ultimately rests on manual aggregation across the daily-bucketed leaderboard — a process vulnerable to double-counting/undercounting the long tail of hundreds of smaller models not shown in the top-10, a technicality that could easily shave the total below the 550T line even if top-named-model growth looks robust. BOTTOM LINE: The freshest data point available (DeepSeek's first negative weekly growth reading, reported Oct 1) plus a stalled weekly-tracker figure unchanged across two consecutive builds spanning the market's open, several top-10 models already shrinking month-over-month, and a 28-day resolution window that is structurally shorter than the 30-day "This Month" metric most extrapolations rely on, together support a case that October's total lands at or below the 550T line rather than comfortably above it. NOT FOUND: No confirmed weekly OpenRouter token figure for the week of Sep 28, 2026 (the first week fully inside the market window) was available in any source as of this research; no official single-number OpenRouter monthly aggregate (for September or any prior month) exists to validate extrapolation methods against a true settled baseline.
3. Forecasters 5 draws · Opus 5.5 + GPT-5.6 Sol · plain mean · re-drawn after crux research
crux research
The first round disagreed: the draws' mean 0.69 sits 0.23 from the same-question market "this kalshi market" at 0.46. One more research loop went after the crux and every draw below was redone with its report.
first-round draws: 71%, 71%, 76%, 64%, 62% · crux loop: 4 searches, 2 page reads
report: crux
CRUX: Whether the most recent confirmed weekly OpenRouter run-rate (~145.8T tokens, week of Sep 21–27) is being sustained or fading into October — i.e., is the Sep 21 surge (driven largely by the Space Bunny Alpha stealth model, Hy4 preview, DeepSeek V4.1 Flash) a durable new baseline or a one-off spike. - **2026-09-21 to 27 (confirmed week):** Total platform volume reached 146 trillion tokens, up 13.18% week-over-week , with Chinese AI large models collectively contributing 62.22 trillion tokens, despite a 7.77% week-over-week decline, still overwhelmingly exceeding the US's 14.2 trillion tokens . Space Bunny Alpha debuted at 13.9 trillion tokens in weekly call volume, trailing only DeepSeek V4.1 Flash and Zhipu GLM 5.3 Flash . - **Live rankings page (data through Sep 30, 2026, fetched Oct 2):** The "This Week" leaderboard (freshest available, covering roughly Sep 24–30) shows **Space Bunny Alpha surging to 28.4T tokens (>999% change)**, DeepSeek V4.1 Flash at 22.7T (+23%), while GLM 5.3 Flash fell to 10.6T (-44%) and Hy4 preview fell to 7.48T (-42%). Source: openrouter.ai/rankings (fetched directly). - Summing just the top-10 models on that live weekly table gives **~107T tokens for top-10 alone** — before adding the long tail of hundreds of smaller models, which historically adds a substantial share. - The "Today" (daily) table on the same live page sums to **~16.7T/day across the top 10**, closely matching forecasters' independent estimate, implying a daily run-rate consistent with (not below) the prior week's pace. - DeepSeek-specific data confirms some softening: OpenRouter's Sep 21 week data showed a strong 24.4% share of requests but also a 2% week-over-week decline for DeepSeek, while the negative growth figure was the primary catalyst for traders pulling back from bullish expectations for the current week — but this is share/request based, not total-tokens based, and other labs (Google +12%, OpenAI +6%, Xiaomi +32%) picked up share in the same period, net not clearly reducing total volume. - Net picture: some individual launch-spike models (GLM, Hy4, DeepSeek share) are cooling as forecasters' "NO" scenario anticipated, but this is being offset or more than offset by continued growth in Space Bunny Alpha and DeepSeek V4.1 Flash, so the aggregate does not show a clear rollback toward the earlier ~127–129T/week baseline. - No single authoritative aggregate "total weekly tokens" figure for the week of Sep 28–Oct 4 (the first week fully inside the market's window) was found as of Oct 2; OpenRouter's own rankings page still only shows data through Sep 30. NOT FOUND: An explicit, sourced total-platform weekly token figure for the week of Sep 28–Oct 4 (first full week of the market window) or any newer aggregate beyond the Sep 21–27 week's confirmed 146T total; no confirmation of whether OpenRouter computes the "October" aggregate as a simple sum of daily/weekly buckets or via some other methodology. BOTTOM LINE: The freshest available data (through Sep 30) shows mixed but on-net still-elevated usage — individual spike-driven models like GLM and Hy4 are fading as the "NO" case anticipated, but this is offset by continued strong growth in Space Bunny Alpha and DeepSeek V4.1 Flash, with top-10 daily volume (~16.7T/day) still tracking near or above the ~19.6T/day breakeven pace once the long tail is added. This modestly favors the "YES" (above 550T) side of the fundamentals debate, though not decisively, since no confirmed full-week aggregate exists yet for any week inside the actual market window.
claude-opus-5-5
50% conf
Yes 72%
No 28%
(0) Not met yet: window Sep 28–Oct 25 just started; OpenRouter rankings only through Sep 30–Oct 1 with no window aggregate. (a) ~23 days remain; resolution observed 10 AM ET Oct 26. (b) Status quo: 550T/28 days needs ~19.6T/day (~137.5T/week). Last confirmed week (Sep 21–27) was ~146T, after 126.8T and 128.9T; the week through Sep 30 shows top-10 models alone at ~107T, implying ~150–165T/week with the long tail. Flat run-rate gives ~580–640T, a ~5–15% cushion. (c) NO: stealth model Space Bunny Alpha (28.4T/week, ~5T/day) is withdrawn or ends free access, and launch-spike models keep fading (GLM -44%, Hy4 -42% WoW, DeepSeek share -2%), dragging the weekly average under ~137T; plus resolution-source aggregation risk. (d) YES: trend growth continues (series ~4x May→Sep), DeepSeek V4.1 Flash keeps growing, Oct 1 launches add volume, landing ~600T+. Market anchor: live 46% on a 27/99 spread with a 97% last close is thin and noisy (mid ~63%), so I weight fundamentals more, discounted for stealth-model and spike-reversal risk.
claude-opus-5-5
50% conf
Yes 72%
No 28%
(0) Not resolved: rankings data runs only through Sep 30 and the window is Sep 28–Oct 25. (a) About 23 days remain; the value is read Oct 26. (b) The threshold needs ~137.5T/week (19.6T/day). The last confirmed week (Sep 21–27) was ~146T, up 13% from ~129T, so a flat run-rate gives ~584T, i.e. Yes, and the series has compounded fast since May. The Kalshi live 46% sits on a 27/99 spread (prior close 97%), so the anchor is weak and fundamentals get more weight. (c) No: spike models fade (Space Bunny Alpha at 28.4T could drop if its stealth/free period ends; GLM -44% and Hy4 -42% already falling; DeepSeek share dip), pulling weeks under ~137T, or Kalshi's long-tail aggregation undercounts. (d) Yes: the run-rate holds or grows; top-10 daily volume is already ~16.7T before the long tail, DeepSeek V4.1 Flash is growing, new models launched Oct 1, and stealth launches have added 10T+ within days. Key unknowns are the Sep 28 week aggregate and the exact methodology, which keep this well short of certain.
claude-opus-5-5
55% conf
Yes 77%
No 23%
(0) Not yet resolved: OpenRouter's rankings page shows data only through Sep 30, and no aggregate for Sep 28–Oct 25 exists. (a) About 23 of 28 days remain; the value is read at 10:00 AM ET on Oct 26. (b) The 550T threshold needs about 137.5T per week. The last confirmed week (Sep 21–27) was about 146T, up 13% week over week after weeks of about 115T, 127T and 129T. Simply holding that pace gives about 584T. The live weekly view through Sep 30 shows Space Bunny Alpha doubling to 28.4T, and top-10 daily volume is about 16.7T before the long tail, so the status quo sits modestly above the line. Growth has been explosive since May (about 25T to about 146T per week). (c) NO: the stealth Space Bunny Alpha model ends or fades, launch-spike models keep cooling (GLM −44%, Hy4 −42%, DeepSeek request share slipping), weekly volume drops back toward 127–130T, or Kalshi's aggregation method differs from the tracker totals. (d) YES: weekly volume holds at 140–160T or more, helped by Oct 1 launches and broadening vendor growth. The Kalshi live quote of 46% (bid 27, ask 99, last close 97%) is too wide to anchor on. Fundamentals put YES at about three in four, discounted for stealth-model volatility and methodology uncertainty.
gpt-5.6-sol
64% conf
Yes 64%
No 36%
(0) The condition is not already met or impossible: OpenRouter’s rankings only contain data through Sep 30 or Oct 1, and the Sep 28–Oct 25 measurement window must finish and be observed at 10:00 AM ET on Oct 26. (a) As of Oct 2, roughly 23 days remain in the usage window, followed by the specified Oct 26 observation. (b) If the latest confirmed 145.8–146T weekly run-rate persists, four weeks produce about 583T, so the status quo is Yes with a modest cushion over 550T. Fresh OpenRouter data supports some persistence: the latest weekly top ten total about 107T and the daily top ten about 16.7T, while the long tail should lift aggregate usage toward or above the required 19.6T daily pace. (c) No results if launch-driven spikes from Space Bunny Alpha, Hy4, and DeepSeek fade, returning weekly usage below 137.5T, or if the rankings’ aggregation excludes enough long-tail usage to keep the official figure at or below 550T. (d) Yes results if the confirmed 146T baseline holds, with continued Space Bunny Alpha and DeepSeek V4.1 growth offsetting declines in GLM and Hy4. I move above Kalshi’s current 46% Yes anchor to 64% because the freshest primary-source partial data does not show the sharp slowdown implied by that price, although the extremely wide 27%–99% spread, missing first-window weekly aggregate, and resolution-source aggregation ambiguity prevent a stronger forecast.
gpt-5.6-sol
61% conf
Yes 64%
No 36%
(0) The condition is not already resolved: OpenRouter’s rankings only show data through Sep 30 or Oct 1, while the specified Sep 28–Oct 25 window must be observed at 10:00 AM ET on Oct 26. (a) Roughly 23 usage days remain before the Oct 25 cutoff, so substantial uncertainty remains despite several elapsed days. (b) If the latest confirmed 145.8–146T weekly run-rate persists, the 28-day total would be about 583T and resolve Yes, modestly above the 550T threshold. (c) No occurs if launch-driven usage fades toward the earlier 127–129T weekly baseline, particularly as GLM 5.3 Flash and Hy4 preview have recently declined, or if the unreported long tail is smaller than assumed. (d) Yes occurs if Space Bunny Alpha, DeepSeek V4.1 Flash, and the broader model mix sustain the current pace; OpenRouter’s freshest tables show about 107T weekly and 16.7T daily from the top ten alone, with additional long-tail usage. I anchor on Kalshi’s live 46% Yes price but move upward because the confirmed 146T week and fresh partial rankings imply an aggregate pace above the required 19.6T per day, while noting the extremely wide 27%–99% spread makes the displayed price a weak consensus signal. Resolution-source aggregation ambiguity and the absence of a complete week inside the actual window keep this well below certainty.
4. Final plain mean of 5 draws · second round
70% (56% confidence)
Yes: 70% No: 30%
Mean of 5 draws (0.72, 0.72, 0.77, 0.64, 0.64) after one crux research pass. (0) Not met yet: window Sep 28–Oct 25 just started; OpenRouter rankings only through Sep 30–Oct 1 with no window aggregate. (a) ~23 days remain; resolution observed 10 AM ET Oct 26. (b) Status quo: 550T/28 days needs ~19.6T/day (~137.5T/week). Last confirmed week (Sep 21–27) was ~146T, after 126.8T and 128.9T; the week through Sep 30 shows top-10 models alone at ~107T, implying ~150–165T/week with the long tail. Flat run-rate gives ~580–640T, a ~5–15% cushion. (c) NO: stealth model Space Bunny Alpha (28.4T/week, ~5T/day) is withdrawn or ends free access, and launch-spike models keep fading (GLM -44%, Hy4 -42% WoW, DeepSeek share -2%), dragging the weekly average under ~137T; plus resolution-source aggregation risk. (d) YES: trend growth continues (series ~4x May→Sep), DeepSeek V4.1 Flash keeps growing, Oct 1 launches add volume, landing ~600T+. Market anchor: live 46% on a 27/99 spread with a 97% last close is thin and noisy (mid ~63%), so I weight fundamentals more, discounted for stealth-model and spike-reversal risk.
Pipeline Timing
Total pipeline time: 194.0s
Per-tool research timings shown in the Research section above.