ValueRank

Frontier AI model ranking focused on production value

Version: v1.3.1
Updated: July 28, 2026
Scope: 17 DeepSWE Best models (ranked), 12 scored dimensions

What ValueRank is

ValueRank is a production-weighted ranking of frontier AI models. It combines benchmark quality, reliability, speed, and a composite cost term into a single rank-based score.

v1.3.1 ranks n=17 models from the current DeepSWE Best roster, keeps the zero-gap rule (no neutral-50 fills), drops dimensions without full AA coverage on the new models, and restores AA+DeepSWE composite Cost. Kimi K2.7 Code remains on DeepSWE but is excluded from the ranked cohort because AA Intelligence Index total eval cost is unpublished (search evidence).

• Keeps the ranked pool limited to models on the current DeepSWE Best leaderboard with complete Cost inputs
• Explicitly excludes off-roster historical models: Grok-Build-0.1, Gemini 3 Flash, Claude Opus 4.6
• Excludes Kimi K2.7 Code from ranking (no AA Index total eval cost; DeepSWE row preserved in raw-data)
• Includes Grok 4.5, Claude Opus 5, GPT-5.6 Sol/Terra/Luna, Kimi K3, Muse Spark 1.1, and other July 2026 DeepSWE entrants
• Uses only benchmarks with full coverage across all 17 ranked models
• Drops IFBench, Terminal-Bench Hard, and τ²-Bench Telecom (missing on newest AA model pages)
• Reruns the excluded-benchmark audit against official sources first, then secondary implementations

v1.3.1 Ranking

Cost is the average of normalized AA Index total eval cost and normalized DeepSWE avg cost per task (0–100, higher = costlier), then rank-normalized.

Rank	Model	Score	Quality	Composite Cost
1	Grok 4.5	60.6	44.9	10.27
2	Kimi K3	59.2	66.4	30.45
3	Muse Spark 1.1	58.9	39.5	9.34
4	GPT-5.6 Terra	58.8	60.0	27.67
5	GPT-5.6 Sol	58.2	73.8	46.46
6	Claude Opus 5	56.5	74.3	56.48
7	GPT-5.6 Luna	53.9	41.1	14.13
8	GPT-5.5	53.0	60.3	38.36
9	Gemini 3.6 Flash	52.6	36.2	13.14
10	GLM-5.2	52.0	41.2	14.22
11	Gemini 3.1 Pro	51.0	45.6	25.19
12	Claude Fable 5	50.0	73.6	90.97
13	GPT-5.4	47.0	45.6	30.11
14	Gemini 3.5 Flash	42.8	30.5	23.14
15	Claude Opus 4.8	40.8	54.2	58.36
16	Claude Sonnet 5	32.9	45.5	85.61
17	Claude Sonnet 4.6	21.7	17.2	40.26

v1.3.1 Frontier

Undominated on composite cost vs. quality:

• Claude Opus 5
• GPT-5.6 Sol
• GPT-5.6 Terra
• Kimi K3
• Grok 4.5
• Muse Spark 1.1
• Gemini 3.1 Pro

Sources Used in v1.3.1

• DeepSWE (Best roster; updated July 25, 2026)
• Artificial Analysis model pages
• Evidence ledger: .refresh/v1.3/

Files

• scores.md: final rankings and normalized scores
• raw-data.md: benchmark inputs and official-first exclusion audit
• methodology.md: scoring method, weights, cohort rules, and audit policy
• site/index.html: published static site