Xiaomi’s open-source model
Self-reported scores beat GPT-6 Sol
MiMo-V2.6-Pro released this week
DeepSWE 72.6%
Grok 4.7 71.0%
GPT-6 Sol 68.8%
All are scores run by the vendors themselves—so take them with a grain of salt first
Price is the real focus
Input 0.435 USD per million tokens
Output 0.87 USD
Sol is 2 USD and 10 USD
Output is cheaper by more than ten times
MIT license—weights released directly
Context length: 1 million
In plain terms
Closed-source models cut their own prices on one hand
while on the other hand getting compared using open-source models
Next up: reputation on real agent tasks
Self-reported scores beat GPT-6 Sol
MiMo-V2.6-Pro released this week
DeepSWE 72.6%
Grok 4.7 71.0%
GPT-6 Sol 68.8%
All are scores run by the vendors themselves—so take them with a grain of salt first
Price is the real focus
Input 0.435 USD per million tokens
Output 0.87 USD
Sol is 2 USD and 10 USD
Output is cheaper by more than ten times
MIT license—weights released directly
Context length: 1 million
In plain terms
Closed-source models cut their own prices on one hand
while on the other hand getting compared using open-source models
Next up: reputation on real agent tasks