The RemKey Index
LLMs ranked by capability per dollar
Every model trades capability against cost. This index ranks the market on the axis that decides a production bill: capability per dollar, the same axis RemKey's classifier routes on. Public list pricing, output-weighted for real agent workloads. Llama 3.3 70B currently leads.
| # | Model | Class | In $/1M | Out $/1M | Context | Capability | Value |
|---|---|---|---|---|---|---|---|
| 🥇 | Llama 3.3 70B Meta |
Open-source | $0.13 | $0.40 | 128K | 77 | 100 |
| 🥈 | Qwen 2.5 72B Alibaba |
Open-source | $0.35 | $0.40 | 128K | 76 | 85 |
| 🥉 | GPT-4o mini OpenAI |
Value | $0.15 | $0.60 | 128K | 74 | 66 |
| 4 | Llama 3.1 405B Meta |
Open-source | $0.80 | $0.80 | 128K | 84 | 45 |
| 5 | DeepSeek V3 DeepSeek |
Open-source | $0.27 | $1.10 | 64K | 81 | 39 |
| 6 | DeepSeek R1 DeepSeek |
Reasoning | $0.55 | $2.19 | 64K | 85 | 21 |
| 7 | Gemini 2.5 Flash Google |
Value | $0.30 | $2.50 | 1M | 78 | 17 |
| 8 | Claude Haiku 4.5 Anthropic |
Value | $0.80 | $4.00 | 200K | 79 | 11 |
| 9 | o3-mini OpenAI |
Reasoning | $1.10 | $4.40 | 200K | 88 | 11 |
| 10 | Mistral Large Mistral |
Open-source | $2.00 | $6.00 | 128K | 80 | 7 |
| 11 | GPT-4o OpenAI |
Frontier | $2.50 | $10.00 | 128K | 90 | 5 |
| 12 | Gemini 2.5 Pro Google |
Frontier | $1.25 | $10.00 | 1M | 91 | 5 |
| 13 | Claude Sonnet 5 Anthropic |
Frontier | $3.00 | $15.00 | 200K | 93 | 3 |
| 14 | Claude Opus 4.8 Anthropic |
Frontier | $15.00 | $75.00 | 200K | 99 | 1 |
How the Value score works
Value = capability ÷ blended price, normalized so the leader scores 100. Blended price weights output tokens 3:1 over input, because agents generate far more than they read, so output pricing dominates the real bill. Capability is an editorial general-purpose rating (0 to 100) held constant across the index. It is deliberately coarse: this board is about the price axis, not a benchmark claim. Pricing is public list, snapshot 2026-07.
Why a gateway cares. No single model wins every task. The top of this board is where cheap, capable work should route, while the frontier models earn their price only on the requests that need them. RemKey's classifier makes that call per request, then proves the cheaper choice held quality before it counts the savings.
Coming next: the daily routed-model board
This index is industry data through our lens. As RemKey traffic grows we will publish the daily most-routed models: which tiers real agent workloads actually land on, aggregated and anonymized. Authority you can quote.
Route on value, start free Compare gateways
Editorial index for orientation, not investment or procurement advice. Prices change; verify against each provider before you build on them.