FrontierRL
0,001
Tokens / second
Stop. Rationing. Tokens.
STAR is the first model with a 32M token context window, able to generate thousands of tokens per second on a single GPU.
Superhuman Speeds
Decoding throughput, tokens per second.
STAR outputs measured on a single NVIDIA 6000 Pro.
FrontierRL STAR 32M context
0t/s
GPT-5.6 1M context
0t/s
Pokee-Isaac 28B 10M context
0t/s
Muse Spark 1.3 1M context
0t/s
Grok 4.7 0.5M context
0t/s
00
At Budget-Level Prices
How far does $100 go?
FrontierRL STAR $0.40 / M
0M
Pokee-Isaac 28B $1 / M
0M
Muse Spark 1.3 $4.25 / M
0M
Grok 4.7 $6 / M
0M
GPT-5.6 Sol $30 / M
0M
00
Why STAR is different
Not another wrapper.
A new architecture, built for long context.
Actually parse video, audio, text, data all in parallel.
Programs
MIT TNT Research 
Together.ai Accelerator