AgenticEvolve: Evolutionary Program Search with Persistent Language Model Agents that Design and Run Experiments

Ze-Yu Peng; Qing-Yu Lan; Zhan-He Wang; Jun Zhang; Yun-Song Piao

cs.NEOct 22, 2026

AgenticEvolve: Evolutionary Program Search with Persistent Language Model Agents that Design and Run Experiments

Authors: Ze-Yu Peng, Qing-Yu Lan, Zhan-He Wang, Jun Zhang, Yun-Song Piao

Organizations: School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China. · International Centre for Theoretical Physics Asia-Pacific, University of Chinese Academy of Sciences, Beijing 100190, China. · Taiji Laboratory for Gravitational Wave Universe, University of Chinese Academy of Sciences, Beijing 100049, China. · School of Fundamental Physics and Mathematical Sciences, Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences, Hangzhou 310024, China. ·

Abstract

Evolutionary program search with large language models keeps the structure of an evolutionary algorithm and uses the model as its variation operator. Typically, the model generates each candidate program in a single request from parent programs and their scores. We present AgenticEvolve, an evolutionary program search framework in which each island is run by a persistent large language model agent. Between evaluations, the agent designs and runs experiments on the candidate program. To measure the contribution of experimentation, we compare AgenticEvolve with No-analysis, an otherwise identical configuration with experiments disabled. On two heuristic design tasks with three large language models, AgenticEvolve produces better programs than No-analysis and OpenEvolve on both hidden and unseen instances. It reaches the baselines' final quality with far fewer evaluations and comparable cost. On twelve equation discovery problems, it recovers the ground-truth structure of more equations than No-analysis and existing language-model and genetic programming baselines.

Figures & tables

Fig. 1: AgenticEvolve. Left: one island, magnified. A persistent agent edits program.py and commits it for formal evaluation. Between commits, it may write and execute analysis.py to run experiments. No-analysis removes these experiments. The feedback on each commit includes a summary of the archive, and the agent can read any archived program, including those of other islands. Right: islands run concurrently and exchange migration candidates, which appear in the archive summary. The agent decides whether to use them.

Model	Input	Cached input	Output	Reasoning
DeepSeek-V4.1-Flash	0.15	0.003	0.60	high
GLM-5.3-Flash	0.15	0.03	0.50	high
Qwen3.8-Flash	0.15	0.016	0.47	medium

TABLE I: LLMs Used in the Experiments, with List Prices in USD per Million Tokens and Reasoning Effort

Fig. 2: Final program quality on the hidden evaluation set and on unseen instances. The unseen OBP instances are OR1 and fresh sequences from the training distribution, and the unseen DJSS instances are the same scenarios with fresh seeds. Each point is one run. Black squares and bars are the mean

\pm

SD over five runs. Gray dashes mark Best Fit, the FunSearch heuristic and rule R1.

Online bin packing (unseen: OR1)
Model	Method	Hidden	Unseen	Cost ($)	$p$
DeepSeek	AgenticEvolve	3.42 $\pm$ 0.25	2.92 $\pm$ 0.48	1.08	–
	AgenticEvolve (No-analysis)	5.54 $\pm$ 1.05	5.12 $\pm$ 0.92	0.42	$<$ 0.01
	OpenEvolve	5.74 $\pm$ 0.41	5.12 $\pm$ 0.42	3.06	$<$ 0.01
GLM	AgenticEvolve	4.46 $\pm$ 0.80	4.14 $\pm$ 0.54	4.48	–
	AgenticEvolve (No-analysis)	6.30 $\pm$ 0.19	5.61 $\pm$ 0.26	0.56	$<$ 0.01

TABLE II: Heuristic design results: mean

\pm

SD over five runs of the final program’s quality on the hidden set and on unseen instances (lower is better), and estimated cost per run.

p

: two-sided Mann–Whitney test of hidden-set quality against AgenticEvolve.

Fig. 3: Best-so-far quality against formal evaluations. Thin lines are individual runs. Bold lines and bands are the mean

\pm

SD over the common evaluation range. The initial job-shop program scores 1.648 and lies above the plotted range.

Fig. 4: Best-so-far quality against estimated cost. Thin lines are individual runs. Bold lines and bands are the mean

\pm

SD over the common evaluation range. Circles mark each run’s actual spend. The initial job-shop program lies above the plotted range.

Fig. 5: Estimated final waste and the optimized potential function, plotted against remaining capacity

L

, for the first commit of island 0 in GLM run 1 on OBP. Circles: the estimate from the agent’s seventh analysis on self-generated sequences. Line: the table

U

used by the committed program after coordinate descent. The dotted line marks the smallest item size, 20.

Method	Acc ID	Acc OOD	Eval.	Requests	Cost ($)
AgenticEvolve	12	12	316	4939	6.54
AgenticEvolve (No-analysis)	8	8	3988	7809	12.12
LLM-SR	3	3	12000	12000	111.34
SR-Scientist	7	8	83	5496	6.59
PySR	1	1	–	–	–

TABLE III: Symbolic regression results, totals over the twelve problems. Acc ID and Acc OOD count the problems whose selected equation is within tolerance

10^{-3}

on the in-domain and out-of-domain test sets. For SR-Scientist, evaluations are the calls of its evaluator tool.

Problem	AgenticEvolve	AgenticEvolve (No-analysis)	LLM-SR	SR-Scientist	PySR
CRK21	$\bullet$ 6.8e-15	– 2.0e-05	– 1.2e-05	– 5.6e-07	– 4.8e-05
CRK35	$\circ$ 3.0e-14	$\bullet$ 2.9e-14	– 9.2e-05	$\bullet$ 2.9e-14	– 3.8e-03
BPG18	$\bullet$ 1.7e-12	$\bullet$ 1.7e-12	$\circ$ 1.7e-12	$\circ$ 2.2e-14	– 3.4e-03
PO10	$\bullet$ 1.4e-14	– 1.8e-02	– 2.0e-02	$\bullet$ 1.4e-14	– 5.5e-04
PO0	$\circ$ 6.8e-13	– 7.2e-05	– 6.9e-04	– 1.2e-03	– 1.1e-02
PO26	$\bullet$ 1.0e-12	$\circ$ 1.0e-12	– 1.5e-03	– 1.1e-02	– 3.2e-02

TABLE IV: Structure of the selected equation on each problem (

\bullet

exact,

\circ

negligible extra terms, – not recovered) and its normalized mean squared error on the out-of-domain test set, bold when within tolerance

10^{-3}

. The material-science problems are judged against the reduced truth.

Setting	Default	Used	Reason
num_islands	5	3	Same as AgenticEvolve.
feature_dimensions	code length, diversity	task features	Same features as AgenticEvolve.
migration_interval	50	10	Each island advances about 60 generations in a run, so the default would trigger migration at most once.
population_size	1000	50	A run produces 180 programs and would never fill the default population. Most example configurations distributed with OpenEvolve use 40 to 100.
temperature	0.7	1.0	Same as AgenticEvolve.
max_tokens	4096	model maximum	Same as AgenticEvolve.

TABLE S1: OpenEvolve settings that differ from the defaults of version 0.3.2. The evaluator timeout is given for OBP and DJSS.

		Online bin packing			Dynamic job shop
Model	Method	Hidden	OR1	Fresh	Hidden	Fresh
DeepSeek	AgenticEvolve	3.42 $\pm$ 0.25	2.92 $\pm$ 0.48	3.72 $\pm$ 0.32	0.840 $\pm$ 0.063	0.846 $\pm$ 0.072
	AgenticEvolve (No-analysis)	5.54 $\pm$ 1.05	5.12 $\pm$ 0.92	5.69 $\pm$ 0.95	0.970 $\pm$ 0.028	0.974 $\pm$ 0.025
	OpenEvolve	5.74 $\pm$ 0.41	5.12 $\pm$ 0.42	5.85 $\pm$ 0.42	0.911 $\pm$ 0.041	0.923 $\pm$ 0.025
GLM	AgenticEvolve	4.46 $\pm$ 0.80	4.14 $\pm$ 0.54	4.72 $\pm$ 0.74	0.847 $\pm$ 0.039	0.863 $\pm$ 0.039
	AgenticEvolve (No-analysis)	6.30 $\pm$ 0.19	5.61 $\pm$ 0.26	6.37 $\pm$ 0.16	0.928 $\pm$ 0.074	0.952 $\pm$ 0.069

TABLE S2: Final quality on every evaluation set, mean

\pm

SD over five runs (lower is better). OBP quality is the excess over the

L_{2}

bound in percent, and DJSS quality is the tardiness relative to rule R1. Hidden-set values are the scores recorded during the runs. Values on OR1 and on fresh instances come from a separate evaluation of the final programs.

			AgenticEvolve $-$ No-analysis		AgenticEvolve $-$ OpenEvolve		No-analysis $-$ OpenEvolve
Model	Task	Set	Difference	$p$	Difference	$p$	Difference	$p$
DeepSeek	OBP	Hidden	$-$ 2.12 [ $-$ 2.93, $-$ 1.24]	0.008	$-$ 2.32 [ $-$ 2.73, $-$ 1.98]	0.008	$-$ 0.20 [ $-$ 1.14, 0.66]	0.917 a
		OR1	$-$ 2.20 [ $-$ 3.00, $-$ 1.37]	0.008	$-$ 2.20 [ $-$ 2.71, $-$ 1.69]	0.012 a	0.00 [ $-$ 0.84, 0.75]	0.916 a
		Fresh	$-$ 1.97 [ $-$ 2.73, $-$ 1.16]	0.008	$-$ 2.13 [ $-$ 2.58, $-$ 1.75]	0.008	$-$ 0.16 [ $-$ 1.01, 0.62]	1.000
	DJSS	Hidden	$-$ 0.130 [ $-$ 0.175, $-$ 0.072]	0.016	$-$ 0.071 [ $-$ 0.123, $-$ 0.006]	0.095	0.059 [0.022, 0.100]	0.056
		Fresh	$-$ 0.128 [ $-$ 0.176, $-$ 0.061]	0.032	$-$ 0.077 [ $-$ 0.124, $-$ 0.009]	0.151	0.051 [0.025, 0.079]	0.032

TABLE S3: Differences in mean final quality between methods, with bootstrap 95% confidence intervals and two-sided Mann–Whitney

p

values over five runs per method. Each difference is the first method minus the second, so a negative value favors the first method. Units are those of Table S2 . The

p

values are not corrected for multiple comparisons. a Asymptotic approximation because of ties. All other

p

values are exact.

Model	Task	final quality	Evaluations	Cost ($)	cost ($)
		OpenEvolve	AgenticEvolve		OpenEvolve
DeepSeek	OBP	5.74	7	0.15	3.06
GLM	OBP	5.54	17	0.96	0.69
Qwen	OBP	6.13	3	0.08	1.10
DeepSeek	DJSS	0.911	19	0.19	1.84
GLM	DJSS	0.931	26	0.20	0.48

TABLE S4: Formal evaluations and estimated cost at which the mean best-so-far curve of AgenticEvolve in Figs. 3 and 4 of the paper first reaches the mean final quality of OpenEvolve. The two budgets are read from the two curves separately and do not refer to the same point of the search. The last column is the mean cost of a complete OpenEvolve run.

	Analysis	Result
1	Simulated Best Fit and Worst Fit on 400 sequences and recorded each bin’s final remaining capacity.	Best Fit excess 6.46%, compared with 6.55% in the formal evaluation. Bins with less than 20 units remaining account for 58% of the waste.
2	Scored bins using a piecewise-linear table indexed by the capacity remaining after placement. Starting from Best Fit, adjusted 13 knot values by coordinate descent on 200 sequences.	Excess 5.99% on all 400 sequences.
3	Repeated the descent with 20 knots on 400 sequences.	Exceeded the 60-second analysis limit.
4	Vectorized the simulator, repeated the 20-knot descent and tested the result on 400 sequences from a new seed.	5.99% on the training sequences and 6.19% on the new ones, where Best Fit gives 6.59%. The descent made no further progress.
5	Repeated the descent over all 131 table entries.	Exceeded the 60-second limit.
6	Counted the frequencies of the remaining capacities that would result from placing an item in each candidate bin, then repeated the descent on the eight most frequent values.	No change.

TABLE S5: The ten analyses before the first commit of island 0 in AgenticEvolve run 1 with GLM on OBP. Excess is over the

L_{2}

bound on the agent’s own sequences.

Model	Task	Offered candidates read	Commits after a read	Improved island best	Improved run best
DeepSeek	OBP	53 / 154	42	34	6
DeepSeek	DJSS	39 / 119	34	28	4
GLM	OBP	42 / 153	38	29	6
GLM	DJSS	28 / 142	25	18	5
Qwen	OBP	42 / 174	41	37	4
Qwen	DJSS	38 / 167	35	23	4

TABLE S6: Cross-island reads in the AgenticEvolve heuristic runs, five runs per row.

Method	Nodes	Structure	Equation
CRK21 (chemistry), target $\dot{A}$
Ground truth			$-k_{1}Ae^{-k_{2}t}+k_{3}A\log(k_{4}t+1)$
AgenticEvolve	17	$\bullet$	$A(0.08820\log{(t+2.283)}-0.07279-0.9112e^{-0.08901t})$
AgenticEvolve (No-analysis)	30	–	$A(t(t(t(2.600\cdot 10^{-6}-1.696\cdot 10^{-8}t)-0.0001558)+0.004786)-0.04734)(\sqrt{A}-0.1362t+18.25)-6.356\cdot 10^{-5}$
LLM-SR	26	–	$0.2869A^{1.014}-\frac{3.122A^{1.015}e^{-0.08080t}}{A^{0.002530}-0.02826A+1.633}$
SR-Scientist	24	–	$A(0.3531-0.02998e^{-0.6502t}-0.9788e^{-0.09575t}-0.2555e^{-0.02338t})$

TABLE S7: Selected equation of every method on every problem, with its number of nodes and its structure (

\bullet

exact,

\circ

negligible extra terms, – not recovered). Constants are rounded to four significant digits for display. The assessment uses the full-precision values. Equations of the material-science problems are judged against the reduced truth after substituting

T=273+500\varepsilon

Problem	AgenticEvolve		AgenticEvolve (No-analysis)		LLM-SR		SR-Scientist		PySR
	ID	OOD	ID	OOD	ID	OOD	ID	OOD	ID	OOD
CRK21	1.4e-15	6.8e-15	5.9e-09	2.0e-05	3.1e-09	1.2e-05	9.8e-10	5.6e-07	1.1e-07	4.8e-05
CRK35	2.0e-15	3.0e-14	2.0e-15	2.9e-14	2.6e-08	9.2e-05	2.0e-15	2.9e-14	9.6e-06	3.8e-03
BPG18	1.2e-12	1.7e-12	1.2e-12	1.7e-12	1.2e-12	1.7e-12	2.1e-14	2.2e-14	2.8e-03	3.4e-03
PO10	7.4e-15	1.4e-14	1.8e-03	1.8e-02	6.0e-03	2.0e-02	7.4e-15	1.4e-14	9.4e-05	5.5e-04
PO0	1.8e-13	6.8e-13	8.9e-05	7.2e-05	8.8e-04	6.9e-04	1.5e-03	1.2e-03	1.0e-02	1.1e-02

TABLE S8: Numerical accuracy of the selected equation per problem: normalized mean squared error on the in-domain (ID) and out-of-domain (OOD) test sets, computed locally with the shared scorer. Bold: accuracy within tolerance

10^{-3}

on that set.

Method	Requests	Cached input (M)	Uncached input (M)	Output (M)	Reasoning (M)	Cost ($)	Wall time (min)
AgenticEvolve	4939	568.9	2.4	7.4	6.0	6.54	313
AgenticEvolve (No-analysis)	7809	1232.5	5.6	12.6	11.0	12.12	432
LLM-SR	12000	3.2	3.7	184.6	183.3	111.34	4521
SR-Scientist	5496	131.6	2.9	9.6	7.3	6.59	808
PySR	–	–	–	–	–	–	8

TABLE S9: Model usage of the symbolic-regression runs, totals over the twelve problems: completed requests, input tokens served from the provider’s cache, uncached input tokens, output tokens (of which reasoning), estimated cost at the DeepSeek-V4.1-Flash prices of the main text, and end-to-end wall time. Wall time depends on machine load and API latency and is reported for reference only.

Method	Data available to model-written code	Feedback of a failed evaluation
AgenticEvolve (v0.2.0)	training data, in analysis scripts	output and full traceback
AgenticEvolve (No-analysis, v0.2.1)	none	one-line summary
LLM-SR	none	none
SR-Scientist	training data, in its analysis tool	last line of the error output

TABLE S10: Data available to the model-written code of each symbolic-regression method and feedback of a failed evaluation in the reported runs.

LLMs for Evolutionary Optimization

See all 36 papers

SeaEvo: Advancing Algorithm Discovery with Strategy Space Evolution
Apr 27, 2026Sichun Luo, Yi Huang, Haochen Luo +7Automated Algorithm Discovery Evolutionary Optimization
1The University of Hong Kong · 2JIUTIAN Research, China Mobile · 3City University of Hong Kong
FreeEvolve: Learning to Evolve Beyond Fixed Loops
Oct 6, 2026Lecheng Kong, Like Hui, Nikos Kanakaris +3Self-Evolving Agents LLM Agent Self-Improvement
AWS AI Labs · Georgia Institute of Technology
ExpEvolve: Structuring and Reusing Experience for Open-Ended Discovery
Oct 21, 2026Yongqiang Chen, Chenxi Liu, Deyu Zou +4AI Agents for Scientific Discovery Large Language Model-Guided Scientific Discovery
MBZUAI · Carnegie Mellon University · TMLR Group, Hong Kong Baptist University +2

Apr 27, 2026J/K move · Enter open · S save

SeaEvo: Advancing Algorithm Discovery with Strategy Space Evolution

Sichun Luo, Yi Huang, Haochen Luo, Fengyuan Liu+6

1The University of Hong Kong · 2JIUTIAN Research, China Mobile · 3City University of Hong Kong

Automated Algorithm Discovery Evolutionary Optimization Large Language Model-Guided Optimization LLM-Based Program Synthesis LLMs for Evolutionary Optimization

Open paper AskPDF arXiv

Large Language Model (LLM)-guided evolutionary search is increasingly used for automated algorithm discovery, yet most current methods track search progress primarily through executable programs and scalar fitness. Even when natural-language reasoning is used through heuristic descriptions or reflection, it typically remains transient mutation context or unstructured memory, rather than organized as persistent population-level state over strategic directions. As a result, evolutionary search can struggle to distinguish syntactically different implementations of the same idea, preserve lower-fitness but strategically promising directions, or detect when an entire family of strategies has saturated. We introduce \model, a modular strategy-space layer that turns language-level strategic reasoning into first-class population-level evolutionary state in LLM-driven program search. \model represents each candidate program with an explicit natural-language strategy, clusters the archive by strategy semantics, retrieves behaviorally complementary inspirations, and periodically navigates the strategy landscape to avoid saturated directions. Without modifying the underlying evolutionary algorithms, \model improves existing evolutionary backbones across algorithm discovery, systems optimization, and agent-scaffold design tasks in most settings. Across four systems benchmarks, \model achieves a 20.6% average relative improvement, with the best single run on Prism scoring 3

\times

higher. These results suggest that persistent strategy representations provide a practical mechanism for improving the effectiveness and cost-efficiency of LLM-guided evolutionary search, pointing toward compound AI systems whose search capabilities benefit from the structured accumulation and reuse of algorithmic strategies.

Abstract

LLMs for Evolutionary Optimization

Explore similar work

SeaEvo: Advancing Algorithm Discovery with Strategy Space Evolution

FreeEvolve: Learning to Evolve Beyond Fixed Loops

ExpEvolve: Structuring and Reusing Experience for Open-Ended Discovery