Overview›Multi-step Workflow

APIFlow-Bench 1.0 Leaderboard | Multi-step Workflow

How well a model chains dependent API calls end-to-end, carrying identifiers and intermediate results across steps without dropping them. 68 tasks × 24 models in this slice.

Ranking◴ Generated 2026-07-31☑ 8,160 trials▤ 24 models

Pass rate

85.4%

Error Rate

0.0%

Trials

8,160

Tasks

Models

p95 Time

12.6m

p95 Tokens

668,793

Tool Calls Avg

22.7

Rank	Rank Spread	Model	Pass rate ± 90% CI	Pass (no abstain)	Abstain%	p95 Time	p95 Tokens	Avg Tools
1	1 ↔ 5	kimi-k3	96.5% 93.5%–98.8%	96.8% 328/339	0.3% 1/340	26.6m	459,054	16.7
2	1 ↔ 12	deepseek-v4-pro	95.0% 90.6%–98.5%	95.3% 323/339	0.3% 1/340	9.4m	174,636	21.2
3	2 ↔ 13	glm-5p2	94.4% 90.0%–97.9%	94.4% 321/340	0.0% 0/340	20.7m	366,777	16.9
4	2 ↔ 13	claude-opus-4-8	94.1% 90.0%–97.6%	96.4% 320/332	2.4% 8/340	1.7m	125,249	12.5
4	2 ↔ 14	deepseek-v4-flash	94.1% 90.3%–97.4%	94.4% 320/339	0.3% 1/340	2.1m	182,816	21.4
4	2 ↔ 14	claude-sonnet-5	94.1% 89.4%–97.9%	95.2% 320/336	1.2% 4/340	2.5m	243,031	16.2
4	2 ↔ 15	kimi-k2p7-code	94.1% 90.0%–97.6%	95.0% 320/337	0.9% 3/340	6.3m	165,772	15.4
8	2 ↔ 16	kimi-k2p6	93.8% 89.4%–97.9%	94.1% 319/339	0.3% 1/340	4.2m	358,749	18
8	2 ↔ 16	qwen3p6-plus	93.8% 89.4%–97.6%	93.8% 319/340	0.0% 0/340	13.1m	784,420	20.6
10	3 ↔ 16	gpt-5.6-luna	93.5% 88.2%–97.9%	94.1% 318/338	0.6% 2/340	2.7m	67,531	14.2
10	2 ↔ 16	gpt-5.6-sol	93.5% 88.5%–97.9%	94.6% 318/336	1.2% 4/340	6.2m	141,922	15.2
12	3 ↔ 16	gpt-5.5	93.2% 88.2%–97.6%	93.2% 317/340	0.0% 0/340	1.9m	138,655	15.7
13	4 ↔ 17	claude-sonnet-4-6	92.9% 88.5%–97.1%	92.9% 316/340	0.0% 0/340	1.7m	92,601	24.9
14	6 ↔ 18	qwen3p7-plus	92.4% 87.9%–96.2%	93.5% 314/336	1.2% 4/340	9.4m	526,279	13.2
15	7 ↔ 18	gpt-5.6-terra	92.1% 86.5%–96.8%	93.2% 313/336	1.2% 4/340	2.6m	98,677	14.4
16	9 ↔ 19	minimax-m2p7	91.2% 86.5%–95.6%	91.4% 310/339	0.3% 1/340	1.5m	126,568	14.6
16	9 ↔ 18	kimi-k2p5	91.2% 86.5%–95.6%	91.7% 310/338	0.6% 2/340	1.6m	91,841	13.8
18	9 ↔ 19	claude-haiku-4-5-20251001	90.9% 85.6%–95.6%	91.7% 309/337	0.9% 3/340	2.1m	216,994	17.9
19	18 ↔ 21	gpt-oss-120b	86.8% 80.6%–92.9%	87.3% 295/338	0.6% 2/340	1.0m	142,674	14.6
19	18 ↔ 21	gpt-5-mini	86.8% 81.2%–92.1%	91.6% 295/322	5.3% 18/340	1.1m	118,022	12.7
21	19 ↔ 21	gpt-oss-20b	85.9% 79.4%–91.8%	86.4% 292/338	0.6% 2/340	4.4m	173,029	16.1
22	22 ↔ 22	gpt-5.4-mini	76.2% 69.4%–82.4%	79.0% 259/328	3.5% 12/340	24.7s	41,707	12.6
23	23 ↔ 23	minimax-m3	20.9% 15.0%–26.8%	44.7% 71/159	53.2% 181/340	71.8m	17,232,979	182.3
24	24 ↔ 24	claude-fable-5	12.4% 7.1%–18.2%	12.4% 42/340	0.0% 0/340	3.0m	55,662	3

Diagnostics

Loading the full trial dataset for this view…