BREAKING
Thinking Machines Debuts 'Inkling'
0
B
total params
0
B
active per token
0
T
training tokens
Breadth Over Depth
Inkling
Open-weight
●
Text, image, audio input
●
Fine-tune via Tinker
●
1M-token context
Closed models
ChatGPT, Claude
●
One-size-fits-all
●
API-only access
●
Frontier scores
Benchmarks at Max Thinking Effort
SWE-Bench
77.6
VoiceBench
91.4
MMMU Pro
73.5
Terminal
63.8
0
TB
BF16 VRAM
0
GB
NVFP4 VRAM
0
x
H200s needed
Efficiency-First Open AI Bet
AI NEWS BLITZ
Mira Murati's startup releases its first open-weight model under Apache 2.0.