Over time · 2026-10-08 to 2026-10-12

Every run, every kernel.

neo in white, the engine closest to it at the last run dashed. A hollow point was taken on a busy machine.

How these were taken

Every run the ledgers kept of neo against the others, oldest first: nothing on this tab was measured again for it. A flag is a move of neo's time from one run to the next beyond what a re-run moves by itself (5% at a millisecond and over, 10% under it): green and down is quicker, amber and up is slower, with the day it was recorded. Within the band it is noise and is not flagged. A hollow point was recorded on a busy machine.

Against the field, 1,000,000 rows

load the CSV

10.4 → 2.77 ms

3.8× quicker over 11 runs, 10-08 to 10-11 · Polars 15.2 ms at the last

filter, count

0.021 → 0.018 ms

1.2× quicker over 11 runs, 10-08 to 10-11 · Polars 0.179 ms at the last

filter into a new table

1.40 → 0.380 ms

3.7× quicker over 11 runs, 10-08 to 10-11 · Polars 0.889 ms at the last

sort, as a permutation

1.05 → 1.03 ms

level over 11 runs, 10-08 to 10-11 · Polars 6.68 ms at the last

sort into a new table

3.07 → 2.72 ms

1.1× quicker over 11 runs, 10-08 to 10-11 · Polars 12.2 ms at the last

sum by region

0.051 → 0.020 ms

2.5× quicker over 11 runs, 10-08 to 10-11 · Polars 1.55 ms at the last

Every run as a table

load the CSV · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:52 · busy	10.4	17.2	65.1	256	412
2026-10-08 20:08 · busy	8.78	15.6	57.3	241	426
2026-10-10 11:52 · busy	8.31	15.6	49.2	227	389
2026-10-10 12:00	8.89	15.5	50.0	240	343
2026-10-11 10:29	4.31	16.6	54.2	255	412
2026-10-11 16:03	2.95	16.9	49.0	241	383
2026-10-11 17:08	2.89	15.6	51.3	234	378
2026-10-11 17:31	2.92	15.3	51.0	237	375
2026-10-11 21:01	2.83	15.1	53.3	227	356
2026-10-11 21:13	2.93	14.8	51.6	234	354
2026-10-11 21:23	2.77	15.2	51.8	229	359

filter, count · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:52 · busy	0.021	0.198	0.626	0.337	43.7
2026-10-08 20:08 · busy	0.014	0.208	0.622	0.333	40.9
2026-10-10 11:52 · busy	0.014	0.219	0.568	0.327	35.3
2026-10-10 12:00	0.014	0.191	0.572	0.333	41.4
2026-10-11 10:29	0.014	0.201	0.644	0.331	43.1
2026-10-11 16:03	0.014	0.177	0.566	0.315	39.1
2026-10-11 17:08	0.014	0.171	0.553	0.328	35.0
2026-10-11 17:31	0.014	0.178	0.582	0.345	39.1
2026-10-11 21:01	0.017	0.187	0.534	0.328	37.6
2026-10-11 21:13	0.017	0.178	0.554	0.324	34.5
2026-10-11 21:23	0.018	0.179	0.602	0.328	36.9

filter into a new table · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:52 · busy	1.40	1.08	43.8	12.7	24.3
2026-10-08 20:08 · busy	0.415	1.06	41.6	12.5	24.9
2026-10-10 11:52 · busy	0.319	0.938	40.1	12.0	22.1
2026-10-10 12:00	0.295	1.09	40.2	11.9	24.1
2026-10-11 10:29	0.490	1.14	43.3	12.9	24.5
2026-10-11 16:03	0.383	1.06	40.2	11.9	25.6
2026-10-11 17:08	0.333	1.09	40.4	11.9	23.3
2026-10-11 17:31	0.316	1.06	42.2	12.1	22.5
2026-10-11 21:01	0.364	0.908	40.2	11.7	24.2
2026-10-11 21:13	0.319	0.999	39.8	11.7	20.7
2026-10-11 21:23	0.380	0.889	40.3	11.8	22.0

sort, as a permutation · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:52 · busy	1.05	7.34	—	71.7	—
2026-10-08 20:08 · busy	1.07	7.27	—	69.8	—
2026-10-10 11:52 · busy	1.05	6.35	—	65.2	—
2026-10-10 12:00	1.06	6.62	—	65.2	—
2026-10-11 10:29	1.06	7.23	—	71.3	—
2026-10-11 16:03	1.03	7.11	—	65.6	—
2026-10-11 17:08	1.02	6.80	—	66.6	—
2026-10-11 17:31	1.03	7.18	—	66.4	—
2026-10-11 21:01	1.04	6.58	—	64.1	—
2026-10-11 21:13	1.06	6.83	—	64.7	—
2026-10-11 21:23	1.03	6.68	—	64.8	—

sort into a new table · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:52 · busy	3.07	13.7	27.6	142	135
2026-10-08 20:08 · busy	2.98	12.4	25.0	126	124
2026-10-10 11:52 · busy	2.77	10.7	23.2	113	123
2026-10-10 12:00	2.72	11.8	24.8	115	119
2026-10-11 10:29	2.91	13.7	27.6	129	134
2026-10-11 16:03	2.75	12.5	24.2	119	117
2026-10-11 17:08	2.74	11.7	24.8	112	112
2026-10-11 17:31	2.75	12.1	25.5	114	116
2026-10-11 21:01	2.69	11.8	23.2	111	116
2026-10-11 21:13	2.76	12.2	23.9	109	121
2026-10-11 21:23	2.72	12.2	23.4	110	122

sum by region · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:52 · busy	0.051	2.04	2.75	13.2	53.6
2026-10-08 20:08 · busy	0.035	1.55	2.52	12.6	46.5
2026-10-10 11:52 · busy	0.032	1.49	2.28	12.4	45.0
2026-10-10 12:00	0.032	1.61	2.31	12.4	42.1
2026-10-11 10:29	0.035	1.61	2.40	13.5	45.2
2026-10-11 16:03	0.024	1.60	2.28	12.3	43.0
2026-10-11 17:08	0.019	1.41	2.29	12.8	40.2
2026-10-11 17:31	0.019	1.57	2.30	12.4	42.1
2026-10-11 21:01	0.020	1.70	2.30	12.2	40.4
2026-10-11 21:13	0.020	1.52	2.27	12.0	40.9
2026-10-11 21:23	0.020	1.55	2.24	12.2	39.0

Against the field, 10,000,000 rows

load the CSV

78.8 → 21.7 ms

3.6× quicker over 11 runs, 10-08 to 10-11 · Polars 114 ms at the last

filter, count

0.341 → 0.317 ms

1.1× quicker over 11 runs, 10-08 to 10-11 · Polars 0.621 ms at the last

filter into a new table

4.04 → 3.40 ms

1.2× quicker over 11 runs, 10-08 to 10-11 · Polars 9.13 ms at the last

sort, as a permutation

12.9 → 9.39 ms

1.4× quicker over 11 runs, 10-08 to 10-11 · Polars 71.4 ms at the last

sort into a new table

68.3 → 62.6 ms

1.1× quicker over 11 runs, 10-08 to 10-11 · Polars 139 ms at the last

sum by region

0.490 → 0.401 ms

1.2× quicker over 11 runs, 10-08 to 10-11 · Polars 6.03 ms at the last

Every run as a table

load the CSV · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:53 · busy	78.8	348	1,490	2,675	2,651
2026-10-08 20:09 · busy	90.7	130	392	2,522	2,970
2026-10-10 11:52 · busy	85.8	120	195	2,365	1,327
2026-10-10 12:01	81.8	119	204	2,420	1,504
2026-10-11 10:30	34.8	129	215	2,566	1,420
2026-10-11 16:04	23.6	128	209	2,376	1,515
2026-10-11 17:09	23.9	132	220	2,422	1,387
2026-10-11 17:31	24.1	130	199	2,465	1,541
2026-10-11 21:02	21.9	117	200	2,428	1,372
2026-10-11 21:14	22.3	114	214	2,335	1,472
2026-10-11 21:24	21.7	114	213	2,343	1,450

filter, count · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:53 · busy	0.341	1.29	3.83	3.89	79.5
2026-10-08 20:09 · busy	0.324	0.678	4.12	3.72	68.8
2026-10-10 11:52 · busy	0.305	0.616	2.19	3.08	42.5
2026-10-10 12:01	0.315	0.631	2.20	3.08	42.3
2026-10-11 10:30	0.322	0.621	2.32	3.18	47.1
2026-10-11 16:04	0.317	0.617	2.19	3.10	46.6
2026-10-11 17:09	0.322	0.626	2.21	3.08	40.6
2026-10-11 17:31	0.315	0.619	2.20	3.15	45.1
2026-10-11 21:02	0.313	0.596	2.09	3.06	42.5
2026-10-11 21:14	0.312	0.636	2.19	3.06	41.4
2026-10-11 21:24	0.317	0.621	2.13	3.12	39.3

filter into a new table · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:53 · busy	4.04	17.4	948	155	114
2026-10-08 20:09 · busy	3.39	8.57	485	154	102
2026-10-10 11:52 · busy	3.42	8.91	246	140	69.1
2026-10-10 12:01	3.43	7.63	264	136	65.8
2026-10-11 10:30	3.55	7.66	299	152	78.9
2026-10-11 16:04	3.50	7.92	293	140	69.2
2026-10-11 17:09	3.43	8.73	274	139	75.9
2026-10-11 17:31	3.42	6.85	276	141	64.4
2026-10-11 21:02	3.66	7.39	216	142	61.6
2026-10-11 21:14	3.45	7.21	234	141	63.0
2026-10-11 21:24	3.40	9.13	328	139	65.1

sort, as a permutation · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:53 · busy	12.9	153	—	945	—
2026-10-08 20:09 · busy	9.69	99.4	—	870	—
2026-10-10 11:52 · busy	9.19	69.1	—	809	—
2026-10-10 12:01	9.39	70.3	—	804	—
2026-10-11 10:30	9.88	75.6	—	873	—
2026-10-11 16:04	9.66	73.2	—	808	—
2026-10-11 17:09	9.35	69.7	—	813	—
2026-10-11 17:31	9.43	69.8	—	837	—
2026-10-11 21:02	9.19	72.7	—	794	—
2026-10-11 21:14	9.31	71.6	—	787	—
2026-10-11 21:24	9.39	71.4	—	798	—

sort into a new table · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:53 · busy	68.3	812	1,015	2,809	765
2026-10-08 20:09 · busy	64.0	174	208	2,028	896
2026-10-10 11:52 · busy	63.0	132	171	1,928	469
2026-10-10 12:01	63.1	122	190	1,961	475
2026-10-11 10:30	65.6	137	217	2,176	522
2026-10-11 16:04	64.2	132	188	2,056	451
2026-10-11 17:09	63.4	125	176	1,960	504
2026-10-11 17:31	63.2	125	188	2,040	484
2026-10-11 21:02	62.4	137	189	1,912	459
2026-10-11 21:14	62.3	145	179	1,900	467
2026-10-11 21:24	62.6	139	188	1,913	469

sum by region · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 19:53 · busy	0.490	106	30.4	179	128
2026-10-08 20:09 · busy	0.481	10.7	13.7	157	112
2026-10-10 11:52 · busy	0.474	6.69	9.18	137	69.7
2026-10-10 12:01	0.482	6.83	9.24	140	70.1
2026-10-11 10:30	0.501	6.34	9.43	152	82.1
2026-10-11 16:04	0.476	6.04	8.95	139	64.7
2026-10-11 17:09	0.468	6.26	9.04	139	72.7
2026-10-11 17:31	0.475	6.02	9.17	140	76.2
2026-10-11 21:02	0.465	5.95	9.11	141	63.6
2026-10-11 21:14	0.466	5.60	8.94	138	62.3
2026-10-11 21:24	0.401	6.03	8.96	142	66.1

Against the field, 100,000,000 rows

load the CSV

1,465 → 249 ms

5.9× quicker over 10 runs, 10-08 to 10-11 · Polars 1,186 ms at the last

filter, count

3.52 → 2.98 ms

1.2× quicker over 10 runs, 10-08 to 10-11 · Polars 4.45 ms at the last

filter into a new table

48.1 → 34.3 ms

1.4× quicker over 10 runs, 10-08 to 10-11 · Polars 111 ms at the last

sort, as a permutation

150 → 99.6 ms

1.5× quicker over 10 runs, 10-08 to 10-11 · Polars 1,031 ms at the last

sort into a new table

1,483 → 744 ms

2.0× quicker over 10 runs, 10-08 to 10-11 · Polars 1,542 ms at the last

sum by region

5.41 → 4.11 ms

1.3× quicker over 10 runs, 10-08 to 10-11 · Polars 52.8 ms at the last

Every run as a table

load the CSV · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 20:16 · busy	1,465	1,715	3,976	33,844	29,040
2026-10-10 11:58 · busy	909	1,203	2,315	26,501	16,694
2026-10-10 12:08	861	1,227	2,452	27,357	13,784
2026-10-11 10:38	409	1,340	2,638	28,527	14,573
2026-10-11 16:09	299	1,210	2,288	28,081	14,524
2026-10-11 17:15	285	1,203	2,375	26,600	14,249
2026-10-11 17:38	279	1,206	2,356	26,817	16,094
2026-10-11 21:07	244	1,165	2,374	26,868	15,837
2026-10-11 21:20	244	1,132	2,386	26,444	15,666
2026-10-11 21:29	249	1,186	2,388	27,009	15,900

filter, count · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 20:16 · busy	3.52	4.67	21.7	39.1	167
2026-10-10 11:58 · busy	3.01	3.95	18.7	30.6	137
2026-10-10 12:08	2.89	4.31	19.1	30.6	134
2026-10-11 10:38	3.09	4.48	19.7	31.4	136
2026-10-11 16:09	3.13	4.43	18.9	30.4	124
2026-10-11 17:15	2.91	4.15	18.9	30.6	132
2026-10-11 17:38	2.91	4.27	18.4	30.4	132
2026-10-11 21:07	2.92	4.12	18.2	30.4	129
2026-10-11 21:20	3.00	4.34	18.1	30.4	130
2026-10-11 21:29	2.98	4.45	18.4	30.4	131

filter into a new table · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 20:16 · busy	48.1	151	2,438	1,578	1,361
2026-10-10 11:58 · busy	35.4	136	1,680	1,504	691
2026-10-10 12:08	34.3	125	1,677	1,475	716
2026-10-11 10:38	36.5	137	2,192	1,615	719
2026-10-11 16:09	35.6	100.0	1,533	1,531	626
2026-10-11 17:15	34.7	125	1,831	1,549	711
2026-10-11 17:38	34.4	116	1,624	1,431	708
2026-10-11 21:07	34.2	119	1,670	1,459	696
2026-10-11 21:20	33.9	113	1,721	1,474	687
2026-10-11 21:29	34.3	111	1,753	1,541	693

sort, as a permutation · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 20:16 · busy	150	1,560	—	10,093	—
2026-10-10 11:58 · busy	98.9	1,000	—	9,706	—
2026-10-10 12:08	100	1,107	—	10,051	—
2026-10-11 10:38	104	1,144	—	10,835	—
2026-10-11 16:09	102	1,055	—	9,821	—
2026-10-11 17:15	101	1,055	—	9,929	—
2026-10-11 17:38	99.9	1,083	—	9,866	—
2026-10-11 21:07	99.1	983	—	9,615	—
2026-10-11 21:20	99.4	970	—	9,735	—
2026-10-11 21:29	99.6	1,031	—	9,765	—

sort into a new table · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 20:16 · busy	1,483	2,490	4,453	24,679	8,585
2026-10-10 11:58 · busy	752	1,716	3,059	23,394	5,289
2026-10-10 12:08	750	1,677	2,889	23,562	5,211
2026-10-11 10:38	771	1,833	3,595	25,418	5,392
2026-10-11 16:09	759	1,693	2,929	23,451	4,653
2026-10-11 17:15	750	1,658	2,930	23,554	5,383
2026-10-11 17:38	747	1,614	2,647	23,179	5,173
2026-10-11 21:07	745	1,547	2,702	22,706	4,830
2026-10-11 21:20	744	1,522	2,929	23,092	4,834
2026-10-11 21:29	744	1,542	2,875	23,277	4,970

sum by region · ms

run	neo	Polars	DuckDB	pandas	Spark
2026-10-08 20:16 · busy	5.41	85.8	99.8	1,452	803
2026-10-10 11:58 · busy	5.06	55.7	81.3	1,401	477
2026-10-10 12:08	5.08	53.7	82.8	1,415	454
2026-10-11 10:38	4.94	56.8	85.4	1,530	473
2026-10-11 16:09	4.24	51.3	80.7	1,419	401
2026-10-11 17:15	4.21	53.4	82.3	1,444	466
2026-10-11 17:38	4.07	52.5	83.5	1,403	453
2026-10-11 21:07	4.08	52.1	81.1	1,396	414
2026-10-11 21:20	4.10	52.7	82.1	1,399	415
2026-10-11 21:29	4.11	52.8	84.4	1,398	432

Against the ports: Go, Rust, Swift, Zig

The four ports were timed together on one day (2026-10-12); only the cold load was run twice, so it is the one kernel of theirs with a line. Measured once and so not drawn: load, filter, count, sort, as a permutation, sort into a new table, sum by region — the Evals tab has those. The load in process is neo's against the Rust port's, the one port bench_load_profile.sh runs, which it has done since 2026-10-10.

load, the parse loop in process, 1M rows

5.82 → 2.79 ms

2.1× quicker over 6 runs, 10-10 to 10-11 · Rust 3.11 ms at the last

cold load, a fresh process, 1M rows

8.96 → 5.96 ms

1.5× quicker over 2 runs, 10-12 to 10-12 · Zig 4.32 ms at the last

load, the parse loop in process, 10M rows

44.5 → 21.1 ms

2.1× quicker over 6 runs, 10-10 to 10-11 · Rust 32.6 ms at the last

cold load, a fresh process, 10M rows

45.0 → 43.3 ms

level over 2 runs, 10-12 to 10-12 · Zig 38.8 ms at the last

load, the parse loop in process, 100M rows

632 → 500 ms

1.3× quicker over 2 runs, 10-10 to 10-10 · Rust 298 ms at the last

cold load, a fresh process, 100M rows

420 → 426 ms

level over 2 runs, 10-12 to 10-12 · Zig 454 ms at the last

Every run as a table

load, the parse loop in process, 1M rows · ms

run	neo	Go	Rust	Swift	Zig
2026-10-10 16:21 · busy	5.82	—	3.06	—	—
2026-10-10 17:35	5.09	—	2.98	—	—
2026-10-10 18:11 · busy	5.45	—	2.95	—	—
2026-10-10 21:53	4.87	—	3.22	—	—
2026-10-11 08:14 · busy	3.58	—	3.09	—	—
2026-10-11 14:11	2.79	—	3.11	—	—

cold load, a fresh process, 1M rows · ms

run	neo	Go	Rust	Swift	Zig
2026-10-12 06:38	8.96	10.4	7.29	6.69	4.45
2026-10-12 07:17	5.96	9.97	6.26	6.54	4.32

load, the parse loop in process, 10M rows · ms

run	neo	Go	Rust	Swift	Zig
2026-10-10 16:21 · busy	44.5	—	34.4	—	—
2026-10-10 17:35 · busy	41.6	—	38.0	—	—
2026-10-10 18:11 · busy	44.1	—	35.6	—	—
2026-10-10 21:53	36.6	—	36.3	—	—
2026-10-11 08:14 · busy	28.9	—	34.5	—	—
2026-10-11 14:11 · busy	21.1	—	32.6	—	—

cold load, a fresh process, 10M rows · ms

run	neo	Go	Rust	Swift	Zig
2026-10-12 06:44	45.0	79.9	58.3	59.7	39.8
2026-10-12 07:18	43.3	77.3	56.5	55.4	38.8

load, the parse loop in process, 100M rows · ms

run	neo	Go	Rust	Swift	Zig
2026-10-10 16:21 · busy	632	—	367	—	—
2026-10-10 21:54 · busy	500	—	298	—	—

cold load, a fresh process, 100M rows · ms

run	neo	Go	Rust	Swift	Zig
2026-10-12 07:12	420	726	504	482	448
2026-10-12 07:19	426	734	542	496	454

On public data, read as text

worldcitiespop, first load

29.2 → 23.8 ms

1.2× quicker over 4 runs, 10-11 to 10-11 · Arrow 26.3 ms at the last

worldcitiespop, loads after it

19.3 → 11.4 ms

1.7× quicker over 4 runs, 10-11 to 10-11 · Arrow 23.0 ms at the last

NYC taxi · 1M, first load

85.3 → 72.9 ms

1.2× quicker over 5 runs, 10-11 to 10-11 · Arrow 96.1 ms at the last

NYC taxi · 1M, loads after it

71.6 → 52.9 ms

1.4× quicker over 5 runs, 10-11 to 10-11 · Arrow 90.7 ms at the last

NYC taxi · 10M, first load

747 → 448 ms

1.7× quicker over 3 runs, 10-11 to 10-11 · Arrow 954 ms at the last

NYC taxi · 10M, loads after it

710 → 418 ms

1.7× quicker over 3 runs, 10-11 to 10-11 · Arrow 959 ms at the last

ClickBench · 1M, first load

185 → 126 ms

1.5× quicker over 6 runs, 10-11 to 10-11 · Polars 274 ms at the last

ClickBench · 1M, loads after it

164 → 96.0 ms

1.7× quicker over 6 runs, 10-11 to 10-11 · Polars 189 ms at the last

ClickBench · 10M, first load

2,070 → 1,053 ms

2.0× quicker over 3 runs, 10-11 to 10-11 · Polars 2,814 ms at the last

ClickBench · 10M, loads after it

2,045 → 1,056 ms

1.9× quicker over 3 runs, 10-11 to 10-11 · Polars 2,696 ms at the last

Every run as a table

worldcitiespop, first load · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:02	29.2	29.8	51.4	278	1,722
2026-10-11 09:09 · busy	29.8	29.8	52.7	280	1,739
2026-10-11 09:46	25.6	29.6	49.2	280	1,125
2026-10-11 16:12	23.8	26.3	47.5	252	1,072

worldcitiespop, loads after it · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:02	19.3	26.6	42.8	102	1,724
2026-10-11 09:09 · busy	19.5	26.5	43.4	99.2	1,749
2026-10-11 09:46	12.6	26.8	44.0	102	1,108
2026-10-11 16:12	11.4	23.0	39.9	91.8	1,053

NYC taxi · 1M, first load · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:04	85.3	103	144	937	3,665
2026-10-11 09:11	89.5	104	146	908	3,699
2026-10-11 09:26	89.6	104	146	908	3,696
2026-10-11 09:50	78.1	104	146	908	3,305
2026-10-11 16:15	72.9	96.1	142	827	3,151

NYC taxi · 1M, loads after it · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:04	71.6	99.0	105	725	3,679
2026-10-11 09:11	73.0	98.4	103	716	3,707
2026-10-11 09:26	74.0	98.0	109	730	3,658
2026-10-11 09:50	57.2	98.3	107	740	3,201
2026-10-11 16:15	52.9	90.7	95.3	661	3,103

NYC taxi · 10M, first load · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:16 · busy	747	1,011	1,302	2,949	—
2026-10-11 10:07	490	1,021	1,331	3,042	32,853
2026-10-11 16:34	448	954	1,245	2,752	31,837

NYC taxi · 10M, loads after it · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:16 · busy	710	1,011	1,267	2,648	—
2026-10-11 10:07	465	1,018	1,268	2,879	32,863
2026-10-11 16:34	418	959	1,098	2,539	31,512

ClickBench · 1M, first load · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:06 · busy	185	—	291	1,758	6,574
2026-10-11 09:13 · busy	192	403	295	1,742	6,431
2026-10-11 09:30	206	413	301	1,835	6,517
2026-10-11 09:52	141	406	294	1,765	—
2026-10-11 10:27	145	402	299	1,780	6,527
2026-10-11 16:21	126	373	274	1,647	5,944

ClickBench · 1M, loads after it · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:06 · busy	164	—	206	1,518	6,505
2026-10-11 09:13 · busy	166	405	203	1,515	6,421
2026-10-11 09:30	182	419	219	1,541	6,497
2026-10-11 09:52	113	407	208	1,523	—
2026-10-11 10:27	115	406	216	1,528	6,498
2026-10-11 16:21	96.0	373	189	1,438	5,876

ClickBench · 10M, first load · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:22 · busy	2,070	4,049	3,017	4,887	—
2026-10-11 10:19	1,572	4,147	3,042	5,589	—
2026-10-11 16:43	1,053	3,752	2,814	4,626	—

ClickBench · 10M, loads after it · ms

run	neo	Arrow	Polars	DuckDB	pandas
2026-10-11 09:22 · busy	2,045	4,099	2,972	5,153	—
2026-10-11 10:19	1,673	4,159	2,980	5,640	—
2026-10-11 16:43	1,056	3,829	2,696	4,682	—

Measured on Apple M5 Max, 18 cores, 26.4.1. Every point is a run a benchmark script appended to the engine's ledgers; nothing on this page was measured again for it.

All evals Every comparison Start a conversation