AI AGENT STORE · MODEL MOMENTUM INDEX
Most popular models
with Hermes.
Find your next agent model. Compare popularity and recent trends, see the relative cost, and launch with your choice already selected.
Our 0–100 index combines popularity with a recent-trend adjustment of up to 20% in either direction. It reflects the observed sample; it is not a quality benchmark or a count of users. Score differences are index points, not “times more popular.”
DeepSeek V4.1 Flash 1017
67.45 / 100 · Reference cost: 1×
The momentum ranking
- 01
DeepSeek V4.1 Flash 1017
67.45 / 100 - 02
DeepSeek V4.1 Flash (September)
67.32 / 100 - 03
Upstage: Solar Mini 4
49.59 / 100 - 04
Z.ai: GLM 5.3 Flash
44.60 / 100 - 05
MiniMax M3.1 Flash
39.94 / 100 - 06
Step 5 Preview
32.18 / 100 - 07
NVIDIA: Nemotron 3 Ultra
27.26 / 100 - 08
Anthropic: Claude Opus 5.5
26.91 / 100 - 09
Z.ai: GLM 5.3
22.09 / 100 - 10
StepFun: Step 3.7 Flash
20.08 / 100 - 11
Anthropic: Claude Sonnet 5.5
18.18 / 100 - 12
Xiaomi: MiMo-V2.6-Pro
15.62 / 100 - 13
Gemini 4 Argon
13.60 / 100 - 14
MiniMax: MiniMax M3
13.41 / 100
How we calculate the index
MMI = popularity score × (1 + recent-trend adjustment), capped at 100. Popularity is the foundation. Recent growth or decline can adjust it by at most 20%.
Popularity uses the source’s published monthly aggregate. For the trend, the latest three complete UTC days receive 50%, 30% and 20% weight, newest first. We compare that weighted share with the preceding three days using the same weights. The current partial day is excluded from the trend.
Every supported model with a verified monthly popularity observation qualifies, even with limited daily history. If any of the six trend days is missing, its adjustment is neutral and the model stays listed. We include all currently supported models with a published popularity observation.
This ranking uses popularity and trend observations for this harness only. Exact model versions are scored separately. The launch selector also includes all other supported available models.
Formula details
Let s be the model’s share of the published monthly aggregate. Popularity P = 100 × ln(1 + 100s) / ln(101). Let r be the weighted daily share for the latest three complete days and b the same weighted share for the preceding three days. Adjustment a = clamp(r/b − 1, −0.20, +0.20). If b is zero and r is positive, a = +0.20; if both are zero or the six-day trend is incomplete, a = 0. The harness index is min(100, P × (1 + a)), rounded to two decimals. Combined scores average the available harness indices. Equal scores sort by model name.
Prices and choosing a model
Cost comparisons use Platform Credit rates with an illustrative mix of 80% input and 20% output tokens. The first model in this page is the reference. The combined page’s reference can differ from the recommendation for an individual harness. Long-context and time-based price changes can produce a range; the highest known multiplier determines the price category. An unquantified premium rate also requires confirmation.
A higher index does not promise better results, and a higher price does not guarantee a smarter model. Test the tasks that matter to you. Running agents keep their chosen model.
Sources and updates
AI Agent Store calculates this index from public Hermes observations and OpenClaw observations published by OpenRouter. They cover activity visible to that source, excluding direct-provider and local use. The source groups some models into “Others”; we do not treat missing individual figures as zero.
The daily catalog review validates model availability and pricing, then runs the same calculation and publishes the updated snapshot. If validation fails, the previous verified snapshot is retained. Monthly popularity observed: 2026-11-13. Last calculation: .