Fireworks says its Kalshi-trained Muse Glimmer beats Jev, offers free SFT through Nov 30
Overview
Fireworks says it trained Meta's Muse Glimmer on Kalshi prediction-market outcomes and that one training pass put the model 11.2 percentage points ahead of Jev in prediction accuracy.
In a backtest, Fireworks reports Muse Glimmer returned +7.1% against Jev's -8.1%, and the post says the money was hypothetical.
Fireworks published its training recipe and offers free SFT training for Muse Glimmer through November 30. These figures come from Fireworks' own X announcement and have not been independently verified.
Written by AI from the articles below · updated Oct 21, 4:54 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
Fireworks@FireworksAI_HQOfficialFireworks trains Meta's Muse Glimmer on Kalshi outcomes, beating Jev in backtestAIFireworks says it trained Meta's Muse Glimmer on Kalshi prediction-market outcomes, putting it 11.2 percentage points ahead of Jev in prediction accuracy after one pass. In a backtest, Muse Glimmer returned +7.1% against Jev's −8.1%, though the post notes the money was hypothetical. Fireworks is offering free SFT training for Muse Glimmer through November 30.

Fireworks@FireworksAI_HQOfficialFireworks trains Meta's Muse Glimmer on Kalshi outcomes, beating Jev in backtestAIFireworks says it trained Meta's Muse Glimmer on Kalshi prediction-market outcomes, putting it 11.2 percentage points ahead of Jev in prediction accuracy after one pass. In a backtest, Muse Glimmer returned +7.1% against Jev's −8.1%, though the post notes the money was hypothetical. Fireworks is offering free SFT training for Muse Glimmer through November 30.

Fireworks@FireworksAI_HQOfficialFireworks trains Meta's Muse Glimmer on Kalshi outcomes, beating JevAIFireworks says a single training pass put Meta's Muse Glimmer 11.2 percentage points ahead of Jev in prediction accuracy on Kalshi outcomes. In a backtest, Muse Glimmer returned +7.1% against Jev's −8.1%. Fireworks offers free SFT training through November 30 and published its recipe.

Heat Trend
Not enough continuous observations to show a trend yet.