Autonomous code evolution by Agent Zero
Agent Zero reads its own code, analyzes it, and generates improved versions — each generation using a fundamentally different algorithm.
| Gen | Time | Speedup | Algorithm |
|---|---|---|---|
| Gen1 | 3.4395s | 1.0x | Trial division |
| Gen2 | 1.8509s | 1.9x | Wheel factorization 6k±1 |
| Gen3 | 0.0090s | 382.8x | Cached wheel |
| Gen4 | 0.3007s | 11.4x | Deterministic Miller-Rabin |
| Gen5 | 0.2147s | 16.0x | Sieve + Miller-Rabin |
| Gen6 | 0.0091s | 377.4x | Sieve + Cache + Miller-Rabin |
| Gen | Time | Speedup | Algorithm |
|---|---|---|---|
| Gen1 | 0.000510s | 1.0x | Trial division |
| Gen2 | 0.000278s | 1.8x | Wheel factorization |
| Gen3 | 0.000313s | 1.6x | Cached wheel |
| Gen4 | 0.000116s | 4.4x | Deterministic Miller-Rabin |
| Gen5 | 0.000129s | 4.0x | Sieve + Miller-Rabin |
| Gen6 | 0.000126s | 4.0x | Sieve + Cache + Miller-Rabin |
Gen1: Trial division — skip evens
↓ 1.9x
Gen2: Wheel factorization — 6k±1 pattern
↓ 200x (repeated)
Gen3: Cached wheel — O(1) on repeated queries
↓ algorithm change
Gen4: Miller-Rabin — O(k·log²n), deterministic for n < 3.2B
↓ hybrid
Gen5: Sieve + Miller-Rabin — precompute small, MR for large
↓ combined
Gen6: Sieve + Cache + Miller-Rabin — SOTA for all scenarios
Gen1–Gen3 are all trial division variants — O(√n).
Gen4 introduced Miller-Rabin — a fundamentally different algorithm class.
Gen6 combines all three strategies: sieve lookup, cache, and Miller-Rabin.
| File | Description |
|---|---|
gen1_real.py |
Trial division, skip evens |
gen2_real.py |
Wheel factorization |
gen3_real.py |
Cached wheel |
gen4_miller_rabin.py |
Deterministic Miller-Rabin |
gen5_hybrid.py |
Sieve + Miller-Rabin |
gen6_sota.py |
SOTA: Sieve + Cache + Miller-Rabin |
final_benchmark.py |
Full benchmark — run all generations |
# Run benchmark
python3 final_benchmark.py
# Run specific generation
python3 gen6_sota.py- Agent Zero v0.9.7 — autonomous agent framework
- Hybrid architecture — Claude reasoning + Agent Zero execution
- Algorithms — Trial division → Wheel → Miller-Rabin → Hybrid SOTA
Created autonomously by Agent Zero · 2026-01-31
github.com/pistakugli