Small models, built for collective intelligence.
23N Labs builds small language models designed to do more with less—specialized, efficient, and able to work together like a colony.
At 23°N, in Gandhinagar near Ahmedabad, we study intelligence the way nature builds it: through small rules, careful experiments, and many parts working as one.
01 — Thesis
The question
Can a 1B–3B model, purpose-built, hold what a 7B–14B model only carries by default?
Most capability gains of the last three years came from scale — more parameters, more data, more compute. That path works, but it isn't the only one, and it isn't available to most teams or most devices.
We start from the opposite constraint: fix the budget, then ask how much capability can be recovered through distillation, targeted training, and task-specific architecture — before reaching for more parameters.
02 — Philosophy
How we work
Coordination over scale.
No single ant knows the route
A colony finds the shortest path through simple local signals, not central planning. We look for the model-scale equivalent before adding parameters.
Small builder, precise structure
The baya weaver builds one of the most structurally exacting nests in nature at a fraction of a larger bird's size. Precision substitutes for mass.
Claims wait for benchmarks
Nothing here is described as solved. R01 is current focus, not a finished result — we report what's measured, not what's hoped.
03 — R01
Current focus
Compression & distillation from 7B–14B into 1B–3B
Under a fixed 1B–3B parameter budget, how much of a 7B–14B foundation model's task capability can be recovered through distillation and targeted fine-tuning — and where does the gap stay structural rather than closeable?
04 — Careers
Work on the unanswered
Build what does not exist yet.
Research Engineer, Distillation
Design the experiments that make compact models more capable: distillation, evaluation, and training systems with clear evidence behind every claim.
Start a conversation →ML Systems Engineer
Turn research into dependable systems that run within real constraints—latency, memory, connectivity, and the devices people actually use.
Start a conversation →Research Intern
Take one sharp question from paper to prototype. Bring curiosity, strong fundamentals, and the habit of measuring before claiming.
Start a conversation →05 — Collaborate
Work with us
Looking for academic and pilot partners
23N Labs is at prototype stage on R01 and is looking for two kinds of partners:
- Academic collaborators — researchers working on distillation, small language models, or parameter-efficient training who want to compare notes or co-investigate.
- Pilot partners — organizations in manufacturing, healthcare, or government/public sector in Gujarat, or similar constrained-infrastructure contexts, willing to test R01 once benchmarks are validated.