PRIMER
AI that learns
after training.
Today’s AI models are largely frozen once training ends.
Primer is a new class of experimental AI designed to acquire knowledge from experience, retain what it learns, and apply that knowledge later.
PRIMER 1.3B/RESEARCH PREVIEW
Training ends.
Learning shouldn’t.
Modern AI can reason over enormous amounts of knowledge.
But after deployment, its underlying intelligence is largely fixed.
Primer investigates a different question:
Can an AI continue acquiring knowledge after its original training is complete?
Capability
Learn something new.
Then remember it.
Then use it later.
Primer is designed to acquire previously unseen concepts from evidence rather than relying only on information contained in its original training.
Knowledge that survives verification can persist beyond the original interaction and become available to future problems.
Results
Measured after training.
Novel Knowledge Acquisition Rate (NKAR)
NKAR measures Primer’s ability in our experimental environment to correctly acquire rules generated after the model checkpoint is frozen.
0.852
Best demonstrated NKAR
Primer 1.3B · Internal post-freeze research benchmark
Gen I
0.000
Gen II
0.333
Gen III
0.778
Gen IV
0.852
Evaluation targets are generated after checkpoint freeze to reduce the possibility that success comes from memorized training examples.
Internal research benchmark. External validation pending.
Memory
Learning isn’t useful
if you forget.
PR
1.00
Retention
PI
1.00
Interference resistance
14 / 14
Retention and interference probes passed in the M9 experiment.
FDR
0.00
For the tested verified-memory pathway.
In our experimental memory evaluation, Primer retained previously acquired knowledge even after intervening learning tasks.
The goal is persistent intelligence: knowledge acquired today should remain useful tomorrow.
Composition
The next problem is harder.
Can knowledge become a building block?
Remembering is not the same as developing.
A developing intelligence should be able to use something it learned previously to understand something it has never encountered before.
Primer’s current research focuses on this transition:
LEARN
↓
REMEMBER
↓
BUILD
22 / 22
Compositional targets solved when the required prerequisite knowledge was available.
This experiment demonstrated that previously available knowledge could support successful solutions to harder related problems. The remaining research challenge is making that process fully autonomous and general.
1.3B
parameters
Start small.
Understand learning.
Primer intentionally begins with a compact model.
Our central research question is not simply whether adding more parameters produces more capability. It is whether an AI can develop mechanisms for acquiring knowledge after deployment.
Once those mechanisms are understood, scale becomes another experimental variable.
Research program
Progress.
- Foundation
Measuring post-training learning
Completed - Acquisition
Novel knowledge acquisition demonstrated
Completed - Efficiency
Learning efficiency improved
Completed - Memory
Persistent knowledge demonstrated
Completed - Composition
Knowledge reuse demonstrated
Completed - Generalization
Expanding what Primer can learn
In progress - Next
Scaling and external evaluation
Upcoming
Method
Built to be tested.
Post-freeze
Test knowledge the model could not simply have memorized from its original training.
Verified
A generated answer is not automatically treated as knowledge.
Persistent
Learning should survive beyond a single interaction.
Measurable
Positive and negative experimental results both matter.
Positioning
A different axis of intelligence.
Primer is not currently designed to replace frontier language models.
It investigates a complementary capability:
What happens after training?
Model
Primer 1.3B
Primer 1.3B
Research preview- Parameters
- 1.3B
- Focus
- Post-training learning
- Deployment
- Local + Cloud
- Status
- Experimental
- Availability
- Coming soon
Primer 1.3B is the first research release in the Primer model family. The research preview is intended for developers and researchers interested in adaptive AI systems.
Demonstration
Don’t take our word for it.
Teach Primer something.
$ primer --demo
Deterministic demonstration of the interaction pattern. Live research preview access is coming soon.
Research
Research before claims.
Primer is an active experimental research program.
We evaluate capabilities under frozen protocols, preserve negative results, and distinguish internal benchmarks from external validation.
Current research areas
- Post-training learning
- Persistent knowledge
- Knowledge transfer
- Generalization
- Scaling
- Evaluation
What happens when AI
doesn’t stop learning?
PRIMER
AI that learns after training.