Matbench Discovery

Column presets:

Discovery test set:

#	Model	CPS	Acc	F1 ↑	DAF	Prec	MAE	R²	κ_SRME	RMSD	CMDS	CDS	Params	Targets	Date Added	r_cut	Training Set
1	EquiformerV3+DeNS-OAM	0.902	0.978	0.931	6.074	0.928	0.018	0.868	0.118	0.059	n/a	n/a	30.3M	EFS_G	2026-04-07	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
2	TECE-OAM-RRA-1.0	0.908	0.978	0.929	6.073	0.928	0.018	0.871	0.093	0.058	0.599	0.233	222M	EFS_G	2026-07-05	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
3	EquFlashV2	0.907	0.978	0.929	6.069	0.928	0.018	0.873	0.094	0.058	n/a	n/a	44.9M	EFS_G	2026-06-11	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
4	GRACE-3L-OAM-L	0.900	0.977	0.925	6.041	0.923	0.018	0.875	0.121	0.058	0.739	0.699	42.1M	EFS_G	2026-07-02	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
5	eSEN-30M-OAM	0.888	0.977	0.925	6.069	0.928	0.018	0.866	0.17	0.061	0.625	0.648	30.2M	EFS_G	2025-03-17	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
6	PET-OAM-XL	0.898	0.977	0.924	6.075	0.929	0.019	0.864	0.119	0.060	0.634	0.388	730M	EFS_G	2026-01-10	n/a	6.6M (113M) MPtrj+OMat24+sAlex
7	MatRIS-10M-OAM	0.877	0.976	0.921	6.039	0.923	0.019	0.871	0.218	0.060	0.581	0.678	10.4M	EFS_GM	2025-10-29	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
8	EquFlash	0.888	0.975	0.919	5.983	0.915	0.019	0.871	0.158	0.060	n/a	n/a	28.7M	EFS_G	2025-06-23	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
9	eqV2 M	0.558	0.975	0.917	6.047	0.924	0.020	0.848	1.771	0.069	0.595	0.283	86.6M	EFS_D	2024-10-18	12 Å	3.37M (102M) MPtrj+OMat24
10	TACE-OAM-L	0.889	0.972	0.910	5.898	0.902	0.020	0.868	0.126	0.061	0.625	0.749	82.9M	EFS_G	2026-04-09	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
11	Nequip-OAM-XL	0.886	0.971	0.906	5.869	0.897	0.020	0.872	0.125	0.063	0.623	0.796	32.1M	EFS_G	2025-11-30	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
12	SevenNet-Omni-i12*	0.873	0.971	0.906	5.954	0.910	0.021	0.868	0.192	0.062	0.626	0.696	54.9M	EFS_G	2026-01-12	6 Å	243M COSMOSDataset
13	ORB v3	0.860	0.971	0.905	5.912	0.904	0.024	0.821	0.21	0.075	n/a	n/a	25.5M	EFS_G	2025-04-05	6 Å	6.47M (133M) MPtrj+Alex+OMat24
14	AlphaNet-v1-OAM*	0.769	0.968	0.901	5.747	0.879	0.024	0.831	0.643	0.079	0.653	0.739	4.65M	EFS_G	2025-05-12	5 Å	6.6M (113M) MPtrj+OMat24+sAlex
15	Allegro-OAM-L	0.840	0.966	0.895	5.674	0.867	0.022	0.868	0.319	0.065	0.602	0.828	9.7M	EFS_G	2025-09-08	7 Å	6.6M (113M) MPtrj+OMat24+sAlex
16	Nequip-OAM-L	0.870	0.967	0.893	5.823	0.890	0.022	0.865	0.166	0.065	0.670	0.818	9.6M	EFS_G	2025-09-08	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
17	DPA-3.1-3M-FT	0.802	0.963	0.884	5.667	0.866	0.023	0.869	0.469	0.069	0.647	0.533	3.27M	EFS_G	2025-06-05	6 Å	163M OpenLAM
18	GRACE-2L-OAM-L	0.865	0.964	0.883	5.840	0.893	0.022	0.862	0.169	0.064	0.759	0.694	26.4M	EFS_G	2025-09-09	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
19	GRACE-2L-OAM	0.837	0.963	0.880	5.774	0.883	0.023	0.862	0.294	0.067	0.751	0.527	12.6M	EFS_G	2025-02-06	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
20	ORB v2 MPA	0.528	0.965	0.880	6.041	0.924	0.028	0.824	1.734	0.097	0.750	0.678	25.2M	EFS_D	2024-10-11	10 Å	3.25M (32.1M) MPtrj+Alex
21	EquiformerV3+DeNS-MP	0.830	0.956	0.863	5.479	0.838	0.029	0.840	0.275	0.070	n/a	n/a	30.3M	EFS_G	2026-04-07	6 Å	146k (1.58M) MPtrj
22	MatterSim v1 5M	0.767	0.959	0.862	5.852	0.895	0.024	0.863	0.575	0.073	0.659	0.717	4.55M	EFS_G	2024-12-16	5 Å	17M MatterSim
23	DPA-4.0.1-Pro-MPtrj	0.840	0.956	0.857	5.609	0.857	0.029	0.836	0.211	0.069	0.598	n/a	22.8M	EFS_G	2026-06-11	6 Å	146k (1.58M) MPtrj
24	MACE-MPA-0	0.795	0.954	0.852	5.582	0.853	0.028	0.842	0.412	0.073	0.659	0.728	9.06M	EFS_G	2024-12-09	6 Å	3.37M (12M) MPtrj+sAlex
25	MatRIS-10M-MP	0.778	0.951	0.847	5.422	0.829	0.031	0.824	0.489	0.072	0.593	0.264	10.4M	EFS_GM	2025-10-29	6 Å	146k (1.58M) MPtrj
26	eSEN-30M-MP	0.797	0.946	0.831	5.260	0.804	0.033	0.822	0.34	0.075	0.593	0.417	30.1M	EFS_G	2025-03-17	6 Å	146k (1.58M) MPtrj
27	GNoME	n/a	0.948	0.829	5.523	0.844	0.035	0.785	n/a	n/a	n/a	n/a	16.2M	EF_G	2024-02-03	5 Å	6M (89M) GNoME
28	GRACE-1L-OAM	0.761	0.944	0.824	5.255	0.803	0.031	0.842	0.517	0.072	0.749	0.552	3.45M	EFS_G	2025-02-06	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
29	eqV2 S DeNS	0.522	0.939	0.815	5.042	0.771	0.036	0.788	1.676	0.076	0.615	0.258	31.2M	EFS_D	2024-10-18	12 Å	146k (1.58M) MPtrj
30	Eqnorm MPtrj	0.756	0.929	0.786	4.844	0.741	0.040	0.799	0.408	0.084	0.639	0.711	1.31M	EFS_G	2025-05-26	6 Å	146k (1.58M) MPtrj
31	HIENet	0.707	0.929	0.777	4.932	0.754	0.041	0.793	0.642	0.080	0.653	0.649	7.51M	EFS_G	2025-07-01	5 Å	146k (1.58M) MPtrj
32	ORB v2 MPtrj	0.470	0.922	0.765	4.702	0.719	0.045	0.756	1.726	0.101	0.727	0.667	25.2M	EFS_D	2024-10-14	10 Å	146k (1.58M) MPtrj
33	Nequip-MP-L	0.730	0.921	0.761	4.704	0.719	0.043	0.791	0.466	0.086	0.629	0.573	9.6M	EFS_G	2025-09-08	6 Å	146k (1.58M) MPtrj
34	SevenNet-l3i5*	0.714	0.920	0.760	4.629	0.708	0.044	0.776	0.55	0.085	0.651	0.533	1.17M	EFS_G	2024-12-10	5 Å	146k (1.58M) MPtrj
35	Nequix MP	0.729	0.914	0.751	4.455	0.681	0.044	0.782	0.446	0.085	0.603	0.561	708k	EFS_G	2025-08-17	6 Å	146k (1.58M) MPtrj
36	Allegro-MP-L	0.720	0.915	0.751	4.516	0.690	0.044	0.778	0.504	0.082	0.556	0.747	18.7M	EFS_G	2025-09-08	6 Å	146k (1.58M) MPtrj
37	Nequix MP PFT	0.755	0.914	0.748	4.479	0.685	0.044	0.784	0.307	0.087	0.616	0.565	708k	EFSH_G	2026-01-08	6 Å	154k (1.59M) MPtrj+MDR-MP PBE ω_q
38	GRACE-2L-MPtrj	0.681	0.895	0.691	4.163	0.636	0.052	0.741	0.525	0.090	0.698	0.355	15.3M	EFS_G	2024-11-21	6 Å	146k (1.58M) MPtrj
39	MACE-MP-0	0.637	0.878	0.669	3.777	0.577	0.057	0.697	0.682	0.091	0.633	0.627	4.69M	EFS_G	2023-07-14	6 Å	146k (1.58M) MPtrj
40	CHGNet	0.400	0.851	0.613	3.361	0.514	0.063	0.689	1.717	0.095	0.554	0.297	413k	EFS_GM	2023-03-03	5 Å	146k (1.58M) MPtrj
41	M3GNet	0.428	0.812	0.569	2.882	0.441	0.075	0.585	1.409	0.112	n/a	n/a	228k	EFS_G	2022-09-20	5 Å	62.8k (188k) MPF

Download table as Subscribe via RSS

The CPS (Combined Performance Score) is a metric that weights discovery performance (F1), geometry optimization quality (RMSD), and thermal conductivity prediction accuracy (κ_SRME). Use the radar chart to adjust the importance of each component.

The training set column shows the number of materials used to train the model. For models trained on DFT relaxations, we show the number of distinct frames in parentheses. In cases where only the number of frames is known, we report the number of frames as the training set size. (N=x) in the Model Params column shows the number of estimators if an ensemble was used. DAF = Discovery Acceleration Factor measures how many more stable materials a model finds compared to random selection from the test set. The unique structure prototypes in the WBM test set have a 15.3% rate of stable crystals, meaning the max possible DAF is

(32.9k / 215k)^−1
            ≈ 6.54

CPS

CPS Progress Over Time

Each point is a model placed at its benchmark inclusion date; the dashed step line traces the running best ("SOTA frontier") CPS v1, so its jumps mark the models that set a new record when they joined the leaderboard. Use the axis/color/size selectors to compare models across any pair of metrics and parameters. The plot shows the same model cohort as the metrics table above, following the active task preset and table filters.

Marker Size

Params 41 models

Log Scale Size

Matbench Discovery is an interactive leaderboard that ranks ML interatomic potentials across crystal stability prediction, geometry optimization, phonons and thermal conductivity, molecular dynamics, and diatomic potential-energy curves.

We rank 41 models covering multiple methodologies including graph neural network (GNN) interatomic potentials, GNN one-shot predictors, iterative Bayesian optimizers and random forests with shallow-learning structure fingerprints.

EquiformerV3+DeNS-OAM leads the Discovery view with the best F1 of 0.931 .

The benchmark exposes accuracy, robustness, and computational-cost trade-offs across these tasks to help users choose models for static and finite-temperature materials simulations.

📖 Important: In Matbench Discovery, the convex hull used to evaluate stability is constructed from DFT reference energies, not from model predictions. This differs from some other benchmarking approaches and has important implications for metric interpretation. See /tasks/discovery for more information.

To cite Matbench Discovery, use:

Riebesell, J., Goodall, R.E.A., Benner, P. et al. A framework to evaluate machine learning crystal stability predictions. Nat Mach Intell 7, 836–847 (2025). https://doi.org/10.1038/s42256-025-01055-1

We welcome new models additions to the leaderboard through GitHub PRs. See the contributing guide for details and ask support questions via GitHub discussion.

Disclaimer: We evaluate how accurately ML models predict several material properties like thermodynamic stability, thermal conductivity, and atomic positions, in all cases using PBE DFT as reference data. Although these properties are important for high-throughput materials discovery, the ranking cannot give a complete picture of a model’s overall ability to drive materials research. A high ranking does not constitute endorsement by the Materials Project.

🆕 New task — Molecular Dynamics. Matbench Discovery now scores how faithfully MLIPs reproduce finite-temperature observables of ab-initio MD (AIMD): radial distribution functions, vibrational density of states, pressure distributions, and single-point energy/force RMSEs. Explore the new metrics on the MD leaderboard.

⚠️ Interpret with caution. The molecular dynamics task is preliminary. The DynaMat v1.0 reference test set and the metrics are still evolving. Treat the current MD metrics and ranking as indicative only, expect changes as test set and metrics evolve.

The reference set currently holds 17 structures (DynaMat v1.0, spanning pure metals, alloys, high-entropy alloys, transition-metal dichalcogenides, perovskites and molecular crystals at 293–1500 K); an upcoming v2 release will grow this AIMD test set to ~100 structures. Collabs to grow it even further welcome! The public reference data intentionally omits energies and forces. Energy/force RMSEs shown here are maintainer-computed private-label diagnostics and are excluded from CMDS, which ranks trajectory-level observables plus speed. All models currently on the leaderboard were run through a unified script, models/run_md.py. If your model isn’t listed, we invite you to run it and submit your metrics via PR.

For details on the MD modeling task, the DynaMat reference set and the CMDS metric, refer to arXiv:2607.03433.

GitHub Activity

Development activity and community engagement of MLIP GitHub repos. Points are sized by number of contributors and colored by number of commits over the last year.