Overview›Discovery

APIFlow-Bench 1.0 Leaderboard | Discovery

Whether a model can find the right API surface on its own — locating endpoints, operations, and parameters from the spec and docs instead of being handed them. 73 tasks × 24 models in this slice.

Ranking◴ Generated 2026-07-31☑ 8,760 trials▤ 24 models

Pass rate

72.1%

Error Rate

0.0%

Trials

8,760

Tasks

Models

p95 Time

8.2m

p95 Tokens

222,112

Tool Calls Avg

17.1

Rank	Rank Spread	Model	Pass rate ± 90% CI	Pass (no abstain)	Abstain%	p95 Time	p95 Tokens	Avg Tools
1	1 ↔ 6	claude-sonnet-4-6	82.5% 75.3%–89.3%	82.5% 301/365	0.0% 0/365	52.8s	62,369	9.3
1	1 ↔ 6	claude-opus-4-8	82.5% 75.1%–89.3%	82.9% 301/363	0.5% 2/365	1.1m	94,890	10.4
3	2 ↔ 14	glm-5p2	79.5% 72.1%–86.3%	79.5% 290/365	0.0% 0/365	5.6m	114,702	13.3
4	2 ↔ 15	gpt-5.5	78.9% 71.2%–86.3%	78.9% 288/365	0.0% 0/365	1.0m	57,235	11.5
5	2 ↔ 15	deepseek-v4-flash	78.6% 71.2%–86.0%	78.6% 287/365	0.0% 0/365	1.6m	145,002	15.3
6	3 ↔ 17	kimi-k2p5	77.8% 69.6%–85.2%	77.8% 284/365	0.0% 0/365	1.3m	84,062	10.2
6	3 ↔ 17	qwen3p6-plus	77.8% 69.3%–85.2%	77.8% 284/365	0.0% 0/365	1.6m	105,289	10.9
6	3 ↔ 17	kimi-k3	77.8% 69.6%–85.2%	78.0% 284/364	0.3% 1/365	13.5m	105,215	12.3
9	3 ↔ 18	gpt-5.6-luna	77.5% 69.3%–85.2%	77.5% 283/365	0.0% 0/365	2.1m	47,812	11.6
9	3 ↔ 18	kimi-k2p6	77.5% 69.3%–85.2%	77.5% 283/365	0.0% 0/365	2.4m	234,263	13.3
9	3 ↔ 18	gpt-5.6-sol	77.5% 69.9%–85.5%	77.5% 283/365	0.0% 0/365	3.3m	46,936	12.2
9	3 ↔ 17	deepseek-v4-pro	77.5% 69.3%–84.9%	77.5% 283/365	0.0% 0/365	5.3m	128,679	16.7
13	3 ↔ 18	claude-haiku-4-5-20251001	77.3% 69.3%–84.7%	77.3% 282/365	0.0% 0/365	53.3s	114,813	12.8
13	4 ↔ 18	kimi-k2p7-code	77.3% 69.0%–85.2%	77.3% 282/365	0.0% 0/365	3.4m	71,005	11
15	4 ↔ 18	minimax-m2p7	76.7% 67.9%–84.4%	76.9% 280/364	0.3% 1/365	37.6s	87,002	11.3
16	5 ↔ 19	gpt-5.6-terra	76.4% 68.5%–84.4%	76.4% 279/365	0.0% 0/365	1.8m	57,234	12
17	5 ↔ 19	claude-sonnet-5	76.2% 67.9%–84.1%	76.2% 278/365	0.0% 0/365	1.6m	167,399	12.9
18	8 ↔ 19	qwen3p7-plus	75.1% 67.1%–82.7%	75.3% 274/364	0.3% 1/365	9.5m	572,018	9.6
19	14 ↔ 20	gpt-oss-120b	72.3% 64.4%–80.3%	72.3% 264/365	0.0% 0/365	40.6s	100,743	12.5
20	17 ↔ 21	gpt-5-mini	70.4% 61.9%–78.1%	74.3% 257/346	5.2% 19/365	57.1s	75,510	10.3
21	20 ↔ 22	gpt-oss-20b	67.4% 59.5%–75.3%	68.3% 246/360	1.4% 5/365	3.8m	121,648	13.9
22	21 ↔ 22	gpt-5.4-mini	62.7% 54.5%–70.7%	64.3% 229/356	2.5% 9/365	21.6s	38,081	10.2
23	23 ↔ 23	minimax-m3	30.7% 25.2%–36.4%	44.4% 112/252	31.0% 113/365	61.4m	15,108,649	142.5
24	24 ↔ 24	claude-fable-5	23.3% 17.0%–29.9%	23.3% 85/365	0.0% 0/365	3.6m	69,988	4.2

Diagnostics

Loading the full trial dataset for this view…