Terminal AI plans compared

FROGLM Subscription Scoring Methodology

• Terminal-ready plans only: FROGLM compares subscriptions you can use from the terminal. A plan is listed only if the subscription itself (not a separate pay-per-token API key) includes an official terminal agent or CLI from the provider (e.g. Claude Code, Codex CLI, GitHub Copilot CLI, Cursor CLI, Antigravity CLI, Grok Build, Mistral Vibe, Devin CLI) or an officially supported subscription login in third-party terminal agents such as OpenCode. App-, web- and office-only assistants are excluded. ChatGPT Go qualifies through Codex, which OpenAI includes in Go (the Codex CLI is highlighted from Plus up). Criteria checked against official documentation in October 2026.
• CSVI (Composite Subscription Value Index): Multiplies composite intelligence (IQ_comp = live Intelligence Index × 1.10 for reasoning models) by usage volume (M_vol), agentic surfaces (M_agentic) and ecosystem extras (M_eco), divided by a sublinear cost curve ((Price + 10) / 30)0.75.
• One rubric for every plan: M_vol = 1 + 0.15·log₂(k), where k is the included usage relative to the provider's standard tier (published allowances such as Claude Max 5x/20x, Copilot AI credits, Google AI Ultra 5x/20x or OpenCode Go limits; the price ratio when a provider does not publish one). M_agentic = 1.20 for the terminal agent + 0.10 each for an IDE agent, cloud/background agents and autonomous web research. M_eco = 1.00 + 0.05 each for multi-lab models, subscription login in third-party agents, bundled perks and image/video generation. The reference model is the strongest model the plan's terminal agent can use within its included usage.
• Dynamic Model Benchmark Synchronization: Whenever FROGLM's hourly scraper refreshes models.json from Artificial Analysis, subscription reference models (e.g. Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra, Grok 4.7, GPT-6 Luna) automatically update their benchmark indices and value scores live.
• Usage Caps & Volume Considerations: Entry-level tiers (ChatGPT Go) offer affordable access but throttle volume tightly. Developer tools (Cursor Pro, Devin Desktop Pro, GitHub Copilot) provide high-throughput code autocomplete and autonomous multi-file refactoring agents.
• Prices reflect official US base rates (subject to local tax/currency differences) and are compared on monthly billing; prices reviewed August 2026, reference-model mappings re-verified against official plan pages in October 2026.