TokenWave AI

AI-powered workflows
for every industry

Organizational software work

Can a coding agent take over a teammate’s work? Recover requirements, implement changes, and verify delivery.

#	MODEL	F2P (%)	P2P (%)	TASK (%)
1	Claude-Opus-5.5CoWork · 101 tasks	63.84	93.24	6.93
2	GPT-6 AstraCoWork · 101 tasks	62.91	89.15	7.92
3	Seed EvolvingCoWork · 101 tasks	39.09	94.48	0.99
4	Qwen3.8-MaxCoWork · 101 tasks	35.70	91.80	0.00
5	GLM 5.3CoWork · 101 tasks	31.27	87.72	0.99
6	Qwen3.8-FlashCoWork · 101 tasks	30.70	93.40	0.00
7	DeepSeek V4.1 FlashCoWork · 101 tasks	28.20	94.30	0.00
8	Kimi K3CoWork · 101 tasks	27.08	84.88	0.99
9	GLM 5.2CoWork · 101 tasks	24.40	95.20	0.00
10	Seed 2.1 ProCoWork · 101 tasks	21.01	81.85	0.00
11	DeepSeek V4 ProCoWork · 101 tasks	13.04	90.21	0.00

Autonomous software testing

Can coding agents work as software test engineers? Generate tests, locate defects, and uncover distinct bugs.

MODEL / HARNESS	HP ↑	F2P ↑	EC ↑	CFG ↑	Tests / task
GPT-5.6-SolCodex · Closed-source	86.02	34.09	72.41	81.41	32.91
DeepSeek-V4.1-FlashClaude Code · Open-source	92.25	27.69	71.29	76.89	86.88
GPT-6-AstraCodex · Closed-source	90.37	27.56	81.41	87.01	162.62
Claude-Opus-5Claude Code · Closed-source	93.16	27.07	79.49	83.39	197.24
Kimi-K3Claude Code · Open-source	93.48	25.00	77.53	81.20	120.50
Claude-Sonnet-5Claude Code · Closed-source	90.65	21.78	68.23	72.88	68.74
GLM-5.3Claude Code · Open-source	94.40	21.33	76.64	80.58	142.02

MODEL / HARNESS	File LR	File LP	Func. LR	Func. LP	Hunk LR	Hunk LP
GPT-6-AstraCodex · Closed-source	85.85	71.59	60.56	47.72	59.72	63.27
Claude-Opus-5Claude Code · Closed-source	80.80	78.85	56.55	54.63	55.11	71.90
Kimi-K3Claude Code · Open-source	80.64	76.26	51.60	54.94	51.98	69.72
GPT-5.6-SolCodex · Closed-source	79.80	75.26	52.59	58.55	52.22	70.34
GLM-5.3Claude Code · Open-source	79.78	75.42	49.39	54.92	51.57	68.73
DeepSeek-V4.1-FlashClaude Code · Open-source	76.97	76.09	48.26	55.15	51.36	71.20
Claude-Sonnet-5Claude Code · Closed-source	62.89	73.82	40.08	50.49	41.20	66.33

MODEL / HARNESS	Overall ↑	Single-file	Multi-file
GPT-6-AstraCodex · Closed-source	49.04	59.05	7.84
Claude-Opus-5Claude Code · Closed-source	45.37	54.57	7.84
GPT-5.6-SolCodex · Closed-source	43.30	52.38	5.88
Kimi-K3Claude Code · Open-source	42.72	52.14	3.92
GLM-5.3Claude Code · Open-source	42.34	51.90	2.94
DeepSeek-V4.1-FlashClaude Code · Open-source	37.93	46.19	3.92
Claude-Sonnet-5Claude Code · Closed-source	33.14	40.63	2.00