Solan can now stop and think.
Reasoning effort is a new setting in Solan, starting with Fresco 1.3. It lets the model think before it replies, and you choose how long. More thinking gets more hard questions right, for a little more per reply.
Right answers against cost
Share of our test questions answered correctly at each level, against the estimated cost of one reply.
One question, five levels
We asked Fresco 1.3 the same counting question at every level. Pick a level to see what it replied, and how much of its thinking it used.
Correct answer
Pick the effort that fits
Low and Med are live today. High, XHigh and Max arrive tomorrow as we expand compute, and we'll post in announcements when they're live.
| Level | Good for | Right answers | Thinking, on average | Estimated cost per reply |
|---|
Every question, every level
The test mixes logic, arithmetic and counting. The last two questions need careful counting, which is where thinking time matters most.
A filled circle means every run was right, an empty ring means none were, and a part-filled circle shows the share that were.
In chat and in the API
In chat
Choose a level from the effort menu, next to the model menu. Your choice is remembered on this device.
In the API
Add reasoning_effort to a Fresco 1.4 request. It accepts "low" (the default) and "medium" today.
curl https://api.sennoric.com/v1/chat/completions \
-H "Authorization: Bearer sennoric-sk-…" \
-H "Content-Type: application/json" \
-d '{
"model": "fresco-1.4",
"reasoning_effort": "medium",
"messages": [{"role": "user", "content": "Is 391 prime?"}]
}'
How we tested
The questions
Scoring
A reply counts as right when the first number it commits to is the correct answer. We read every reply that was marked wrong or looked unclear.
Cost
What it doesn't show
These are short, hard questions picked to need thinking. For everyday chat, Low is usually enough, and higher levels also take longer to reply.
Give Solan time to think.
Reasoning effort is live in chat and the API for Fresco 1.4.