CLI + IDE Coding Agents — Plan Pricing vs Usage Limits

CLI + IDE Coding Agents — Plan Pricing vs Token/Context & Usage Limits

Snapshot date: 2026-02-14

Comparing AI coding agents is messy because most vendors do not publish token caps per plan, especially for consumer-oriented CLI and IDE add‑ons. This page distills what is actually disclosed, flags what is not stated, and links to vendor primary sources so you can cite them in acquisition paperwork or budget briefings. Use it alongside the Claude Code Guide, Gemini Code Guide, OpenHands Guide, and the Full-Stack Development with AI workflow for deeper operational context.

How to read this guide

Plan allowance unit: Vendors express limits as tokens, requests, credits, or raw dollars. Pay attention to the unit before comparing plans.
Context window vs. quota: Some vendors publish context window sizes (max tokens in a single prompt) but not recurring monthly allowances.
“Not published” ≠ unlimited: When a doc says limits are not published, that only means the vendor hasn’t shared the numbers publicly.
Shared pools: Several tools share quotas between CLI/IDE agents and their main chat apps. If you burn through Claude Code, you also reduce Claude desktop/web usage.

1) CLI agents

1.1 Claude Code (Anthropic)

Plan inclusion: Claude Code ships with Claude Pro and Claude Max subscriptions and shares usage limits with Claude web/desktop/mobile.
Source: Anthropic Support
Token behavior: Anthropic confirms Claude Code consumes tokens and offers cost guidance, but does not publish a consumer-facing token cap table.
Source: Claude Code – Costs
Context window (consumer): Anthropic describes automatic context management for paid Claude plans but does not state a fixed window size in the help article.
Source: How large is the context window on paid Claude plans?
Enterprise/API note: The Claude API exposes beta 1,000,000 token windows on Opus 4.6 and Sonnet 4.x for certain org tiers, but this is separate from consumer Claude Code limits.
Source: Build with Claude – Context windows
Pricing: Consumer pricing lives at claude.com/pricing; team/enterprise pricing details are in the Claude Code Funding guide.

Wiki-ready takeaways

✅ Included with Pro/Max and shares limits with the main Claude app
⚠️ No published “X tokens/day” number for consumer CLI usage
✅ API/org tiers offer 1M-token beta windows (separate channel)

1.2 Codex CLI (OpenAI)

Plan inclusion: Codex CLI runs locally and can authenticate via ChatGPT Plus/Pro/Business/Edu/Enterprise or by API key.
Source: Codex CLI docs
Pricing: OpenAI explains that some Codex modes track against ChatGPT plan entitlements, while API-key access is pay-per-token using standard GPT pricing.
Source: Codex pricing
Model guidance: OpenAI recommends gpt-5.3-codex for most CLI work; Codex-tuned models appear in the API catalog.
Source: Codex models and Prompting guide
Context window publication: OpenAI does not publish a CLI-specific context window number. Published limits instead focus on workspace model sizes (e.g., GPT-5.1 Thinking at 196K tokens in ChatGPT Enterprise/Edu).
Source: ChatGPT Enterprise and Edu models & limits
Usage caps: ChatGPT help center articles reference weekly message allocations and higher context limits for GPT-5 Thinking, but no table maps those caps directly to Codex CLI.
Source: ChatGPT usage limits

Wiki-ready takeaways

✅ Codex CLI is bundled across the ChatGPT paid family or can run on an API key
⚠️ Numeric CLI context window caps are not stated publicly
✅ Enterprise/Edu workspaces document GPT model context windows (128K / 196K) but that data must be cited carefully because it is not branded as “Codex CLI limits”

1.3 Gemini CLI (Google Gemini Code Assist)

Context window: Google Cloud’s Gemini Code Assist docs state 1,000,000 token context window for “local codebase awareness.”
Source: Gemini quotas
Quota pool: Gemini CLI and “agent mode” share the same quota pool; a single CLI prompt may call multiple backend requests.
Source: Gemini quotas
Requests per user:
- Standard edition: 120 requests/user/minute, 1,500 requests/user/day
- Enterprise edition: 120 requests/user/minute, 2,000 requests/user/day
  Source: Gemini quotas
Broader API rate limits: Google’s AI for Developers docs expand on RPM/TPM/RPD tiers for Gemini APIs.
Source: Gemini API rate limits
Pricing context: Public pricing is fragmented across consumer “Gemini Advanced/AI Pro” bundles and Google Cloud Standard vs Enterprise editions. For CLI/agent comparisons, the Cloud quota doc is the authoritative table.

Wiki-ready takeaways

✅ 1M token local context is explicitly documented
✅ Requests/day and requests/minute caps are public
✅ CLI + agent mode use the same quota pool, so workflows must budget for both

2) IDE agents (Editor-integrated)

2.1 Cursor

Context windows: Cursor typically runs at 200K token context windows. “Max Mode” expands to each model’s maximum (including 1M-token Gemini 3 Pro).
Sources: Max Mode docs and Cursor models
Usage allowance: Cursor Pro bundles $20/month of “frontier model usage” priced at API rates; you can top up at cost.
Source: June 2025 pricing update
Higher tier: Cursor Ultra costs $200/month and offers “20× more usage than Pro,” again denominated in dollar-equivalent usage pools rather than tokens.
Source: New tier announcement
Token disclosure: Cursor has not published a per-plan “N tokens/month” chart; usage depends on which model you route through Max Mode.

2.2 Windsurf

Credits, not tokens: Windsurf plans grant monthly prompt credits: Free 25, Pro 500, Teams 500/user, Enterprise 1,000/user (and higher upon request).
Source: Windsurf pricing
Consumption model: Credits are deducted per agent interaction, with multipliers based on the chosen model.
Source: Windsurf usage docs
Context window disclosure: Windsurf’s “Fast Context” feature explains retrieval behavior but does not publish a numeric token window per plan.
Source: Fast Context

2.3 GitHub Copilot (Coding agent features)

Agent availability: GitHub’s coding agent capabilities are available on Copilot Pro, Pro+, Business, and Enterprise (depending on rollout).
Source: About the coding agent
Billing unit: Copilot tracks requests and premium requests, not raw tokens. Organizations can allocate premium request budgets per seat.
Sources: Copilot premium requests and Request definition
Token context: GitHub does not publish token window sizes. Practical planning is based on how many premium agent requests are in your monthly allocation.

3) Comparison table

Product	Agent Type	Plan allowance unit	Published allowance	Published context window tokens	Notes
Claude Code	CLI	Shared usage limits (Claude app + CLI)	Included with Claude Pro / Max	Consumer CLI: not published; API beta 1M tokens on Opus/Sonnet tiers	Keep team/enterprise purchasing nuances in Claude Code Funding
Codex CLI	CLI	ChatGPT plan entitlements or API pay-per-token	Included with ChatGPT Plus/Pro/Business/Edu/Enterprise; API billed per token	Not publicly published for the CLI	Workspace docs cite GPT-5.x context sizes (128K / 196K) but not branded as Codex CLI
Gemini CLI (Code Assist)	CLI	Requests per user (shared with agent mode)	Standard: 1,500 req/day & 120 req/min; Enterprise: 2,000 req/day & 120 req/min	1,000,000 tokens (local codebase awareness)	Quotas documented in Google Cloud Gemini guide
Cursor	IDE	Dollar-equivalent usage pool	Pro: $20/mo included; Ultra: 20× Pro	Default 200K; Max Mode up to model max (1M+)	Token allowance varies with chosen model cost
Windsurf	IDE	Credits per month	Free 25; Pro 500; Teams 500/user; Enterprise 1,000/user	Not published (model-dependent)	Credits deducted per agent call with multipliers
GitHub Copilot	IDE/Agent	Requests / premium requests	Premium request budgets per org plan	Not published	Limits measured in request budgets, not tokens

4) Community plan tracker

Source: Maintained by the IrregularChat community. Submit PRs (or sheet updates) when vendors change pricing or limits.

Product	Plan / Price (USD)	Allowance metric	Published context window	Notes	Source
Claude Code	Pro $20/mo, Max $35/mo (consumer)	Shared Claude usage pool across web/desktop/CLI	Consumer CLI: Not published; API beta 1M tokens on Opus/Sonnet tiers	Included with Pro/Max; auto context management	Claude pricing · Context window FAQ
Codex CLI	ChatGPT Plus $20/mo, ChatGPT Pro $50/mo, Business/Edu/Enterprise = contract	ChatGPT plan entitlements or API pay-per-token	CLI-specific cap not published; workspace docs cite GPT-5.1 (128K) / GPT-5.1 Thinking (196K)	Runs locally with ChatGPT auth or API key	Codex pricing · Enterprise limits
Gemini CLI (Code Assist)	Standard & Enterprise seats billed via Google Cloud	Standard: 1,500 req/day + 120 rpm; Enterprise: 2,000 req/day + 120 rpm (shared with agent mode)	1,000,000 tokens local codebase awareness	Requests are pooled between CLI + agent workflows	Gemini quotas
Cursor	Pro $20/mo (includes $20 API usage), Ultra $200/mo (20× usage)	Dollar-based pool spent at API pricing; top-ups sold at cost	Default 200K; Max Mode up to each model max (1M+ when supported)	Usage varies with chosen model mix	June 2025 pricing · New tier
Windsurf	Free, Pro, Teams, Enterprise (see pricing page for current USD rates)	Monthly credits: 25 / 500 / 500 user / 1,000 user	Not published (model-dependent via Fast Context)	Credits burn faster on larger models	Windsurf pricing · Usage docs
GitHub Copilot	Pro/Pro+/Business/Enterprise (per-seat, listed on GitHub pricing)	Requests & premium requests per seat/org budget	Not published	Coding agent availability depends on tier; request budgets throttle heavy workflows	Copilot agent · Premium requests

Implementation tips for teams

Budget translation: Convert each vendor’s allowance (credits, requests, dollar pools) into the workload you care about (e.g., “How many full repo refactors per month?”). Capture those assumptions in your SOP so new operators know the tradeoffs.
Pair with workflow guides: Use this page as a factual reference, then hand teammates to the Full-Stack Development with AI playbook for end-to-end harness design.
Document floor vs. ceiling: Vendors reserve the right to throttle during abuse spikes. Keep the authoritative links above handy when leadership needs proof of the published numbers.
Cross-check quarterly: Pricing and quota disclosures change frequently. Set a reminder to refresh these notes when revisiting procurement or renewing subscriptions.

Sources

Claude Code plan inclusion and usage limits: Anthropic Support
Claude Code context guidance: Anthropic Support
Claude Code cost guidance: Claude Code docs
Claude API context: Build with Claude
Claude pricing: claude.com/pricing
Codex CLI docs: OpenAI Codex CLI
Codex pricing and models: OpenAI Codex pricing · Codex models
Codex prompting: OpenAI cookbook
ChatGPT workspace limits: OpenAI Help Center
ChatGPT usage limits: OpenAI Help Center
Gemini quotas and limits: Google Cloud Gemini docs
Gemini API rate limits: ai.google.dev
Cursor context and pricing: Cursor Max Mode · Cursor models · June 2025 pricing · New tier
Windsurf pricing & usage: Windsurf pricing · Usage documentation · Fast Context
GitHub Copilot agent and billing: About the coding agent · Premium requests · Request types

Need more pricing nuance (especially for Anthropic enterprise seats)? See the Claude Code Funding & Subscriptions guide.