Best Small model out there Butt....
And its a big butt. This model thinks a lot, i mean a looooooottttttt. Every single time, i see it got the answer much earlier in its thinking process but then it tries a diff approch. It tries a diff approach 10 or so times. I mean ok, you can try a different approach once or twice to verify your original approach's answer but doing it 10 times is unnecessary
@Faint6005
thanks for the feedback and for pointing this out!
The model currently runs in high-reasoning mode by default, which means it puts extra effort into reasoning even on simple queries. This helps with complex/ambiguous tasks, but can also lead to a bit more token usage and slightly slower responses. We’re already working on optimizing this trade-off so future releases reduce the overhead while keeping the strong performance.
@Faint6005 thanks for the feedback and for pointing this out!
The model currently runs in high-reasoning mode by default, which means it puts extra effort into reasoning even on simple queries. This helps with complex/ambiguous tasks, but can also lead to a bit more token usage and slightly slower responses. We’re already working on optimizing this trade-off so future releases reduce the overhead while keeping the strong performance.
Ohh.... didn't know it also had reasoning levels like gpt oss
@Faint6005 thanks for the feedback and for pointing this out!
The model currently runs in high-reasoning mode by default, which means it puts extra effort into reasoning even on simple queries. This helps with complex/ambiguous tasks, but can also lead to a bit more token usage and slightly slower responses. We’re already working on optimizing this trade-off so future releases reduce the overhead while keeping the strong performance.Ohh.... didn't know it also had reasoning levels like gpt oss
@Faint6005 Just to clarify, we don’t support configurable reasoning levels like GPT-OSS. The “high-reasoning mode” mentioned is to simply describes the model default behavior, where it allocates more internal reasoning effort to improve robustness and accuracy. Users cannot manually adjust this it's baked into how the model processes queries.
We have added a developer note in our ReadMe as well. This is the initial version, built using only CPT and SFT, no RL leading to model performing extensive reasoning by default, and we are actively working on making it more efficient and concise in the next release.
Glad to hear it! Model is strong, but it's really thinking too much, glad to hear you are working on this aspect 💪🏻
@Faint6005 Just to clarify, we don’t support configurable reasoning levels like GPT-OSS. The “high-reasoning mode” mentioned is to simply describes the model default behavior, where it allocates more internal reasoning effort to improve robustness and accuracy. Users cannot manually adjust this it's baked into how the model processes queries.
We have added a developer note in our ReadMe as well. This is the initial version, built using only CPT and SFT, no RL leading to model performing extensive reasoning by default, and we are actively working on making it more efficient and concise in the next release.
Sorry for jumping to conclusions. And i have one more thing to say: This model isn't good at coding