Finance AI Benchmark

Finance Benchmark for LLMs

Large language models may prove to the service economy what steam power was to industry: a general-purpose technology that sharply lowers the cost of skilled work. Nowhere are the stakes higher than in financial services, among the world's most expensive and information-intensive businesses. Yet capturing that value depends less on finding one universal winner than on routing each task to the model best equipped to perform it. Which LLM should underwrite a borrower, interpret a Basel rule or price a derivative? Finance Benchmark supplies the evidence.

Open leaderboard for AI models on real finance workflows — Basel III, credit underwriting, derivative pricing, and more.

26 models73 v2 tasksTasks passed + Finance Index

v2 is the primary leaderboard. Browse the task catalog for v2 and v3 sets.

Leaderboard

26 models

#1
claude-opus-4-7
anthropic
93.2%
Finance Index
92.8
Consistency
2.8/3
Eval cost
$1.14
Tasks passed
93.2%
#2
deepseek-v4-flash
deepseek
93.2%
Finance Index
94.4
Consistency
2.5/3
Eval cost
$0.02
Tasks passed
93.2%
#3
gpt-5.6-sol
openai
91.8%
Finance Index
93.3
Consistency
2.7/3
Eval cost
$1.08
Tasks passed
91.8%
#4
claude-opus-4-8
anthropic
91.8%
Finance Index
91.7
Consistency
2.7/3
Eval cost
$1.23
Tasks passed
91.8%
#5
gpt-5.6-luna
openai
90.4%
Finance Index
92.2
Consistency
2.6/3
Eval cost
$0.98
Tasks passed
90.4%
#6
grok-3
xai
90.4%
Finance Index
90.6
Consistency
2.6/3
Eval cost
$0.64
Tasks passed
90.4%
#7
claude-fable-5
anthropic
90.4%
Finance Index
92.2
Consistency
2.6/3
Eval cost
$1.95
Tasks passed
90.4%
#8
claude-haiku-4-5-20251001
anthropic
89.0%
Finance Index
82.8
Consistency
2.6/3
Eval cost
$0.78
Tasks passed
89.0%
#9
deepseek-v4-pro
deepseek
89.0%
Finance Index
89.4
Consistency
2.6/3
Eval cost
$0.07
Tasks passed
89.0%
#10
grok-4
xai
89.0%
Finance Index
89.4
Consistency
2.6/3
Eval cost
$0.60
Tasks passed
89.0%
#11
gpt-5.5
openai
89.0%
Finance Index
91.1
Consistency
2.5/3
Eval cost
$0.95
Tasks passed
89.0%
#12
kimi-k3
moonshot
89.0%
Finance Index
87.2
Consistency
2.6/3
Eval cost
$3.13
Tasks passed
89.0%
#13
claude-sonnet-4-6
anthropic
87.7%
Finance Index
88.3
Consistency
2.6/3
Eval cost
$1.05
Tasks passed
87.7%
#14
gpt-5.6-terra
openai
87.7%
Finance Index
88.3
Consistency
2.6/3
Eval cost
$0.52
Tasks passed
87.7%
#15
claude-opus-4-6
anthropic
86.3%
Finance Index
87.2
Consistency
2.6/3
Eval cost
$1.87
Tasks passed
86.3%
#16
gemini-2.5-pro
google
84.9%
Finance Index
80.6
Consistency
2.4/3
Eval cost
$0.75
Tasks passed
84.9%
#17
claude-sonnet-5
anthropic
84.9%
Finance Index
80.0
Consistency
2.4/3
Eval cost
$0.42
Tasks passed
84.9%
#18
grok-4.5
xai
84.9%
Finance Index
85.0
Consistency
2.5/3
Eval cost
$0.33
Tasks passed
84.9%
#19
gemini-2.5-flash
google
84.9%
Finance Index
84.4
Consistency
2.4/3
Eval cost
$0.15
Tasks passed
84.9%
#20
deepseek-reasoner
deepseek
84.9%
Finance Index
84.4
Consistency
2.4/3
Eval cost
$0.02
Tasks passed
84.9%
#21
gemini-3.5-flash
google
83.6%
Finance Index
82.8
Consistency
2.5/3
Eval cost
$0.41
Tasks passed
83.6%
#22
kimi-k2.7-code
moonshot
80.8%
Finance Index
77.2
Consistency
2.2/3
Eval cost
$1.21
Tasks passed
80.8%
#23
gpt-5.3-codex
openai
65.9%
Finance Index
46.9
Consistency
2.0/3
Eval cost
$0.52
Tasks passed
65.9%
#24
gpt-5.4
openai
63.0%
Finance Index
40.6
Consistency
1.9/3
Eval cost
$0.53
Tasks passed
63.0%
#25
gpt-5.5-pro
openai
61.6%
Finance Index
43.9
Consistency
1.8/3
Eval cost
$101.61
Tasks passed
61.6%
#26
deepseek-chat
deepseek
60.3%
Finance Index
39.4
Consistency
1.8/3
Eval cost
$0.02
Tasks passed
60.3%

#	Model	Provider	Tasks passed ↓	Finance Index	Consistency	Eval cost
#1	claude-opus-4-7	anthropic	93.2%86.3%–98.6%	92.8	2.8/3	$1.14
#2	deepseek-v4-flash	deepseek	93.2%87.7%–98.6%	94.4	2.5/3	$0.02
#3	gpt-5.6-sol	openai	91.8%84.9%–97.3%	93.3	2.7/3	$1.08
#4	claude-opus-4-8	anthropic	91.8%84.9%–97.3%	91.7	2.7/3	$1.23
#5	gpt-5.6-luna	openai	90.4%83.6%–95.9%	92.2	2.6/3	$0.98
#6	grok-3	xai	90.4%83.6%–95.9%	90.6	2.6/3	$0.64
#7	claude-fable-5	anthropic	90.4%83.6%–95.9%	92.2	2.6/3	$1.95
#8	claude-haiku-4-5-20251001	anthropic	89.0%80.8%–95.9%	82.8	2.6/3	$0.78
#9	deepseek-v4-pro	deepseek	89.0%80.8%–95.9%	89.4	2.6/3	$0.07
#10	grok-4	xai	89.0%82.2%–95.9%	89.4	2.6/3	$0.60
#11	gpt-5.5	openai	89.0%80.8%–95.9%	91.1	2.5/3	$0.95
#12	kimi-k3	moonshot	89.0%80.8%–95.9%	87.2	2.6/3	$3.13
#13	claude-sonnet-4-6	anthropic	87.7%79.5%–94.5%	88.3	2.6/3	$1.05
#14	gpt-5.6-terra	openai	87.7%79.5%–94.5%	88.3	2.6/3	$0.52
#15	claude-opus-4-6	anthropic	86.3%78.1%–93.2%	87.2	2.6/3	$1.87
#16	gemini-2.5-pro	google	84.9%76.7%–93.2%	80.6	2.4/3	$0.75
#17	claude-sonnet-5	anthropic	84.9%76.7%–93.2%	80.0	2.4/3	$0.42
#18	grok-4.5	xai	84.9%76.7%–93.2%	85.0	2.5/3	$0.33
#19	gemini-2.5-flash	google	84.9%76.7%–93.2%	84.4	2.4/3	$0.15
#20	deepseek-reasoner	deepseek	84.9%76.7%–93.2%	84.4	2.4/3	$0.02
#21	gemini-3.5-flash	google	83.6%74.0%–91.8%	82.8	2.5/3	$0.41
#22	kimi-k2.7-code	moonshot	80.8%71.2%–90.4%	77.2	2.2/3	$1.21
#23	gpt-5.3-codex	openai	65.9%54.9%–75.6%	46.9	2.0/3	$0.52
#24	gpt-5.4	openai	63.0%52.0%–74.0%	40.6	1.9/3	$0.53
#25	gpt-5.5-pro	openai	61.6%50.7%–72.6%	43.9	1.8/3	$101.61
#26	deepseek-chat	deepseek	60.3%49.3%–71.2%	39.4	1.8/3	$0.02

openai/gpt-5.5-pro is omitted from the Time · Cost · Rating plots below: at $101.61 in eval cost it would flatten every other model onto a single edge of the cost axis. It remains in the leaderboard table (Tasks passed 61.6%).

Time · Cost · Rating

Rotating 3D tradeoff: eval wall-clock time, list-price cost ($), and Finance Index. Ideal models sit toward low time, low cost, and high rating. Extreme cost outliers are excluded from this plot (see note above) so the rest of the field stays readable. Hover to pause and inspect.

deepseek
openai
anthropic
xai
moonshot
google

2D tradeoff

Pick a pair of axes. Published models with both values are plotted (cost outliers that would collapse the scale are noted above). Points are colored by provider; hover for the model name and values.

anthropic
deepseek
openai
xai
moonshot
google

Collaborate How Scoring Works

Scores by domain

SliceShow

Tasks passed (%) · top 8 models

Finance Benchmark for LLMs

Leaderboard

Time · Cost · Rating

2D tradeoff

Scores by domain

Quant