Agentic workloads are token-intensive — agents loop, retry, and chain multiple calls. That makes cost-per-token a critical factor alongside raw capability. This ranking divides each model's weighted agentic score (Terminal-Bench 2.0, BrowseComp, OSWorld-Verified) by its output token price. The result shows which models give you the most agent capability per dollar. If you're building production AI agents with budget constraints, this is where you start.
Unless noted otherwise, ranking surfaces on this page use BenchLM's provisional leaderboard lane rather than the stricter sourced-only verified leaderboard.
Bottom line: Agentic tasks are token-intensive — value matters more here than almost anywhere. Gemini 3.1 Flash-Lite leads, with GPT-4o mini offering competitive agentic value.
According to BenchLM.ai, Grok 4.1 Fast leads this ranking with a score of 125.45, followed by DeepSeek V3.2 (121.59) and MiniMax M3 (68.63). There is a significant gap between the leading models and the rest of the field.
The best open-weight option is DeepSeek V3.2 (ranked #2 with a score of 121.59). Open-weight models are highly competitive in this category — self-hosting is a viable alternative to proprietary APIs.
This ranking is based on provisional weighted averages across the scoring benchmarks in agentic tracked by BenchLM.ai. For detailed model profiles, click any model name below. To compare two specific models head-to-head, use the "vs #" links.
Grok 4.1 Fast
xAI · 1M
Score: 62.7 · $0.5/1M
DeepSeek V3.2
DeepSeek · 128K
Score: 51.1 · $0.42/1M
MiniMax M3
MiniMax · 1M
Score: 82.4 · $1.2/1M
Gemini 3.1 Flash-Lite leads agentic value — most agent capability per dollar.
GPT-4o mini strong agentic value in OpenAI's lineup.
Gemini 2.5 Flash good agentic performance at Flash-tier pricing.
Get notified when models move. One email a week with what changed and why.
Free. No spam. Unsubscribe anytime.
The best value model is Grok 4.1 Fast by xAI with a provisional Score/$ ratio of 125.45 (score: 62.7, output: $0.5/1M tokens).
The best open-weight model is DeepSeek V3.2 at position #2.
56 models are included in this ranking.
Value scores divide the weighted agentic score by output token price (per 1M tokens). Higher means more capability per dollar. Models with no listed price are excluded.
Value rankings favor cheap models even if absolute performance is modest. A model scoring half as well at one-tenth the price wins on value — but may not meet your quality bar. Always check raw scores alongside value rankings.
For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.
Free. No spam. Unsubscribe anytime.