Dear Kronos team,
First, thank you for releasing Kronos (mini, small, base) under MIT license
and for publishing such thorough research in AAAI 2026. I have been running
Kronos-base locally on Apple Silicon with the NeoQuasar/Kronos-base weights
for intraday forecasting on NQ and ES futures, and the quality has been
genuinely impressive — the domain-specific tokenizer clearly does what
general-purpose TSFMs cannot.
I am writing to respectfully request access to Kronos-large (499.2M params),
which the README indicates is "unavailable for public use" and gated behind
institutional licensing inquiries.
WHO I AM:
I am Kasper Bredahl, a senior project engineer (maskinmester / senior
project manager) from Denmark. I trade my own capital on NQ/ES futures
through prop-firm challenges (Apex, Lucid) using ICT methodology, and I
have built a personal trading terminal ("KB Terminal") that integrates
Kronos-base into a live forecasting workflow. My use case is:
- Pure research and personal trading — NOT commercial resale
- Not redistributed — the model would run locally on my own hardware
- Not republished — weights stay on my own machines
- Forecasts are only for my own trading decisions
WHY I NEED KRONOS-LARGE:
I currently run Kronos-base with N=50 Monte Carlo samples, lookback=40 bars,
T=0.6, top_p=0.9 on QQQ and SPY as proxies for NQ/ES. The results are
valuable but inherently limited by the 102.3M parameter ceiling. Based on
the Table 6 and Table 8 results in your paper, Kronos-large's 499.2M
parameters materially improve RankIC on short-horizon forecasts — which is
exactly the regime I trade.
WHAT I OFFER IN RETURN:
- Formal citation in any public discussion of my setup (blog posts,
social media, if any)
- Detailed feedback from real-world use — which prompting strategies,
lookback configurations, and sample counts work best on NQ/ES futures
specifically, since my data is from a different market segment than
your primary A-share / CSI300 validation
- Any bugs or edge cases I encounter during production use
- Willingness to sign a non-redistribution / non-commercial-resale
agreement if needed
- Attribution and link to your repository in my terminal's "About" page
TECHNICAL CONTEXT:
- Running on Apple Silicon (M5 Pro, 24GB unified memory, Metal backend)
- Python 3.9 environment with your official model loader from
github.com/shiyu-coder/Kronos
- Integration via HTTP API to my React-based trading terminal
- Production-grade: 5-minute caching, pre-warm on startup, ensemble Monte
Carlo aggregation
- I have also studied your Binary Spherical Quantization tokenizer paper
and the AAAI 2026 submission carefully
REQUEST:
I understand if Kronos-large must remain gated, and I fully respect any
licensing terms you would require. If there is any process — paperwork,
non-commercial agreement, academic verification — I am happy to complete
it. If Kronos-large is simply not available for individual traders, I
understand; in that case, any guidance on optimal Kronos-base configuration
for futures data would already be enormously valuable.
Thank you for your time and for releasing this groundbreaking model family.
Your work is genuinely advancing what's possible for independent traders
with limited resources.
Kind regards,
Kasper Bredahl
Kasperbredahl@hotmail.com
Copenhagen, Denmark
Dear Kronos team,
First, thank you for releasing Kronos (mini, small, base) under MIT license
and for publishing such thorough research in AAAI 2026. I have been running
Kronos-base locally on Apple Silicon with the NeoQuasar/Kronos-base weights
for intraday forecasting on NQ and ES futures, and the quality has been
genuinely impressive — the domain-specific tokenizer clearly does what
general-purpose TSFMs cannot.
I am writing to respectfully request access to Kronos-large (499.2M params),
which the README indicates is "unavailable for public use" and gated behind
institutional licensing inquiries.
WHO I AM:
I am Kasper Bredahl, a senior project engineer (maskinmester / senior
project manager) from Denmark. I trade my own capital on NQ/ES futures
through prop-firm challenges (Apex, Lucid) using ICT methodology, and I
have built a personal trading terminal ("KB Terminal") that integrates
Kronos-base into a live forecasting workflow. My use case is:
WHY I NEED KRONOS-LARGE:
I currently run Kronos-base with N=50 Monte Carlo samples, lookback=40 bars,
T=0.6, top_p=0.9 on QQQ and SPY as proxies for NQ/ES. The results are
valuable but inherently limited by the 102.3M parameter ceiling. Based on
the Table 6 and Table 8 results in your paper, Kronos-large's 499.2M
parameters materially improve RankIC on short-horizon forecasts — which is
exactly the regime I trade.
WHAT I OFFER IN RETURN:
social media, if any)
lookback configurations, and sample counts work best on NQ/ES futures
specifically, since my data is from a different market segment than
your primary A-share / CSI300 validation
agreement if needed
TECHNICAL CONTEXT:
github.com/shiyu-coder/Kronos
Carlo aggregation
and the AAAI 2026 submission carefully
REQUEST:
I understand if Kronos-large must remain gated, and I fully respect any
licensing terms you would require. If there is any process — paperwork,
non-commercial agreement, academic verification — I am happy to complete
it. If Kronos-large is simply not available for individual traders, I
understand; in that case, any guidance on optimal Kronos-base configuration
for futures data would already be enormously valuable.
Thank you for your time and for releasing this groundbreaking model family.
Your work is genuinely advancing what's possible for independent traders
with limited resources.
Kind regards,
Kasper Bredahl
Kasperbredahl@hotmail.com
Copenhagen, Denmark