Proprietary

Vals Index

Updated 5/28/2026

Benchmark consisting of a weighted performance across finance and coding tasks. Showing the potential impact that LLMs can have on the economy.

As of May 28, 2026, Claude Opus 4.8 ranks first on the Vals Index v1.1 with 70.17%, followed by GPT 5.5 (67.62%) and Claude Opus 4.7 (66.10%).

Vals Index
ACCURACY

Vals Index leaderboard

Rank Model Accuracy Cost / Test Input / Output Cost Duration
1 Claude Opus 4.8 70.17% $2.60 $5 / $25 20m39s
2 GPT 5.5 67.62% $4.81 $5 / $30 17m36s
3 Claude Opus 4.7 66.10% $5.97 $5 / $25 12m10s
4 Gemini 3.5 Flash 62.05% $1.47 $1.5 / $9 6m11s
5 Claude Sonnet 4.6 60.30% $2.14 $3 / $15 11m34s
6 Qwen 3.7 Max 57.29% $5.30 $2.5 / $7.5 29m38s
7 DeepSeek V4 56.23% $0.84 $1.32 / $3.96 21m09s
8 Kimi K2.6 55.55% $0.71 $0.95 / $4 19m17s
9 Gemini 3.1 Pro Preview (02/26) 53.42% $1.34 $2 / $12 7m42s
10 GLM 5.1 52.14% $0.85 $1 / $3.2 14m48s
11 GPT 5.4 Mini 51.42% $0.76 $0.75 / $4.5 22m03s
12 Gemini 3 Flash (12/25) 49.31% $0.40 $0.5 / $3 6m09s
13 Qwen 3.6 Plus 48.04% $1.43 $0.5 / $3 14m08s
14 Grok 4.3 46.63% $0.71 $1.25 / $2.5 11m04s
15 GPT 5.4 Nano 46.46% $0.36 $0.2 / $1.25 21m24s
16 MiniMax-M2.7 41.41% $0.72 $0.3 / $1.2 13m08s
17 Claude Haiku 4.5 (Thinking) 40.33% $0.55 $1 / $5 6m18s
18 Grok 4.20 (Reasoning) 39.11% $0.49 $2 / $6 3m15s
19 Gemini 3.1 Flash Lite Preview 35.24% $0.17 $0.25 / $1.5 4m47s

Results

Industry Average Accuracy Comparison

Methodology

This archived version uses Terminal-Bench 2.0 in the coding bucket, before Terminal-Bench 2.1 replaced it in v1.2.

Finance (weight: 2.0): ~$2T contribution to U.S. GDP

Coding (weight: 1.4): ~$1.4T contribution to U.S. GDP

Coding = 0.25 * SWE_Bench + 0.25 * TBench + 0.5 * VibeCodeBench
Vals_Index = (2.0 * AVG(CorpFin, FinanceAgent) + 1.4 * Coding) / 3.4