BackgrindWhat does your AI coding agent really cost?
Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, OpenCode, Grindy — set your hours, model mix and seats, and get a ranked monthly estimate with the assumptions on every card. Pricing verified July 3, 2026. Free, no signup.
Not sure which agent fits you? Take the 30-second quiz →Set your scenario
Your estimated monthly bill
Ranked cheapest first at your settings. Every card shows its route and assumption — the full model is documented below.
Move a slider to get your ranked estimate.
No key, no separate model bill — let the agents grind.
Grindy is a hosted, Claude-powered engine with $50/mo of included usage (as metered in the app). It ships inside Backgrind Pro at $29/mo — overlay, ambient notifications and Live mode included.
Get Backgrind Pro — $29/moThe hidden cost: the hours you sit waiting
The subscription is not the whole bill. Every long agent run you watch instead of working is paid attention. Put a number on it:
Backgrind pings you only when a run needs a decision — overlay toasts on the desktop, Live mode on your phone — so those hours go back to real work. Grind with your agents, not next to them.
Reclaim the wait — Backgrind Plus $9/moSubscription vs API: the breakeven math
Everything above is computed from these documented assumptions — change your mix and the breakeven moves with it:
- One focused agent-hour ≈ 1.2M input + 0.18M output tokens after typical prompt-cache savings.
- Frontier mix at Sonnet 4.6 list: 1.2 × $3 + 0.18 × $15 ≈ $6.30 per agent-hour.
- Fast mix at Haiku 4.5 list: 1.2 × $1 + 0.18 × $5 ≈ $2.10 per agent-hour. The mix slider blends the two.
- A month is 4.33 weeks; multi-seat totals are per-seat × seats.
- Flat subscriptions from Anthropic and OpenAI are assumed to cover ≈ 4× face value in list-price tokens before caps bite; Cursor credits count ≈ 1× face value; Copilot uses its exact credit pools ($15 / $70 / $200); usage past a pool is priced at list rates.
- Breakeven: a $20/mo plan beats raw API above 20 ÷ (4.33 × your hourly burn) hours per week — about 0.7 hrs/week at a pure frontier mix, about 2.2 hrs/week at a pure fast mix. Past a plan capacity, the calculator moves you up a tier or adds metered top-ups automatically.
- Gemini API rates are modeled at ≈ 30% below Claude list; Codex BYO-key at ≈ 5% below.
How much do AI coding agents cost in 2026?
Every number below was verified against the linked official source on July 3, 2026. AI pricing changes monthly — the verification date sits next to each row so you know how fresh it is.
| Option | Free tier | Monthly plans | Metered / API pricing | Watch out for | Verified |
|---|---|---|---|---|---|
| Claude Code Anthropic | No free CLI tier — the API is pay-as-you-go from $0 | Pro $20 · Max 5x $100 · Max 20x $200 | API: Sonnet 4.6 $3 in / $15 out per MTok · Haiku 4.5 $1 / $5. Max plans can buy extra usage at API rates. | A 5-hour rolling window plus weekly caps on every plan — heavy weeks can lock you out until reset. (5-hour caps doubled May 6, 2026; weekly caps +50% through July 13, 2026.) | claude.com/pricing 2026-07-03 |
| OpenAI Codex CLI OpenAI | Codex CLI is free software; ChatGPT Free includes limited Codex use | ChatGPT Go $8 · Plus $20 · Pro 5x $100 · Pro 20x $200 | Token-credit pricing since Apr 2, 2026 — tasks draw credits at API token rates, extra credits purchasable. BYO key: gpt-5.3-codex $1.75 / $14 per MTok. | Plans include token credits, not unlimited runs — one long agent session can drain a Plus month in days. The Pro tiers are rate multipliers, not flat unlimited. | developers.openai.com/codex/pricing 2026-07-03 |
| Cursor Anysphere | Hobby $0 — limited completions and agent requests | Pro $20 · Pro+ $60 · Ultra $200 (Teams $40/user) | Included agent usage is credited at roughly API list prices; past the pool it bills usage-based. | Included usage is close to face value in model credits — agent-heavy weeks exhaust the Pro pool quickly and spill into metered billing. | cursor.com/pricing 2026-07-03 |
| GitHub Copilot GitHub / Microsoft | Copilot Free — limited completions and chat | Pro $10 (incl. $15 credits) · Pro+ $39 (incl. $70) · Max $100 (incl. $200) · Business $19 · Enterprise $39 per user | Usage-based since June 1, 2026: 1 AI credit = $0.01, drawn per token at listed model rates; extra credits at the same price. | Credit burn — frontier models eat the included pool fast. The Pro pool ($15) is roughly 2–3 focused frontier agent-hours before you are buying credits. | github.com/features/copilot/plans 2026-07-03 |
| Gemini CLI Google | Historically 60 req/min and 1,000 req/day free — see the retirement note | Google AI Plus $7.99 · Pro $19.99 · Ultra $100–$200 — no longer served by Gemini CLI for individuals | BYO Gemini API key still works; Gemini list prices sit below Claude/OpenAI at comparable tiers. | Retired for individuals: since June 18, 2026 Gemini CLI stopped serving the individual, Google AI Pro and Ultra tiers — Google directs users to Antigravity CLI. BYO API key is the practical path. | developers.google.com — Code Assist quotas 2026-07-03 |
| OpenCode + your API key OpenCode (open source) | Free, open source — the software costs $0 | None required — bring your own API key (75+ providers) | You pay the token list prices of whichever provider you pick. Nothing else. | No caps from the tool itself — your API budget is the only ceiling, so set provider spend limits before a long run surprises you. | opencode.ai 2026-07-03 |
| Grindy — hosted, in Backgrind Pro Backgrind | In-app demo (no persistence) | Included in Backgrind Pro — $29/mo or $290/yr | $50/mo of included Grindy usage (as metered in the app) — a hosted, Claude-powered engine with no API key and no separate model bill; aggressive compaction stretches the included usage. | A single monthly allowance ceiling — heavy frontier use past the allowance throttles until next month instead of surprise-billing you. | backgrind.com/#pricing 2026-07-03 |
Sources — last verified July 3, 2026
- Claude Code: claude.com/pricing · verified 2026-07-03
- OpenAI Codex CLI: developers.openai.com/codex/pricing · verified 2026-07-03
- Cursor: cursor.com/pricing · verified 2026-07-03
- GitHub Copilot: github.com/features/copilot/plans · verified 2026-07-03
- Gemini CLI: developers.google.com — Code Assist quotas · verified 2026-07-03
- OpenCode + your API key: opencode.ai · verified 2026-07-03
- Grindy — hosted, in Backgrind Pro: backgrind.com/#pricing · verified 2026-07-03
- Anthropic — API pricing (Sonnet 4.6, Haiku 4.5) · verified 2026-07-03
- GitHub — Copilot is moving to usage-based billing (June 1, 2026) · verified 2026-07-03
- OpenAI — Codex rate card (token credits, Apr 2, 2026) · verified 2026-07-03
- OpenAI — ChatGPT plans (Go, Plus, Pro 5x / 20x) · verified 2026-07-03
- Google — Gemini Code Assist quotas (June 18, 2026 retirement notice) · verified 2026-07-03
Spotted a stale number? The vendors move fast — tell us at hello@backgrind.com and we will re-verify the row.
AI coding agent cost FAQ
Is Claude Code cheaper on the Pro plan or the API?
Below roughly one hour of agentic coding per week at frontier prices, pay-as-you-go API is cheaper than the $20 Pro plan. Above that, Pro wins fast — a focused agent-hour burns about $6.30 of Sonnet 4.6 tokens at list price, so Pro pays for itself in three or four hours of use per month. Heavy daily users outgrow the 5-hour window and weekly caps and land on Max 5x ($100) or Max 20x ($200), which can still buy extra usage at API rates.
How do GitHub Copilot AI credits work after June 2026?
Since June 1, 2026 Copilot bills by usage: each plan includes a monthly pool of AI credits (1 credit = $0.01) and every request consumes credits based on real token usage at listed model rates. Pro is $10/mo with $15 of credits, Pro+ is $39 with $70, Max is $100 with $200. When the pool runs out you buy more credits at the same rate — so heavy agentic use is effectively metered API pricing with a small included buffer.
Is OpenCode really free?
The software is free and open source — you only pay for the model tokens you use through your own API key, across 75+ providers. At Claude list prices that is about $6.30 per focused agent-hour on Sonnet 4.6 or $2.10 on Haiku 4.5. If you would rather not manage keys at all, Grindy runs OpenCode hosted with a Claude-powered engine inside Backgrind Pro.
What does Grindy cost?
Grindy is included in Backgrind Pro at $29/mo (or $290/yr): a hosted, Claude-powered OpenCode engine with $50/mo of included Grindy usage (as metered in the app) — no API key and no separate model bill, with aggressive context compaction stretching the included usage further than the raw number suggests. Backgrind Plus ($9/mo) is the bring-your-own-agent tier.
How much does Cursor cost per month in 2026?
Hobby is free, Pro is $20/mo, Pro+ is $60, Ultra is $200, and Teams is $40 per user. Included agent usage is credited at roughly API list prices, so agent-heavy weeks can exhaust the Pro pool quickly and spill into usage-based billing — which is why Cursor often ranks mid-pack in this calculator once hours climb.
What happened to the Gemini CLI free tier?
Google retired Gemini CLI for individual users: since June 18, 2026, Gemini CLI and the Code Assist IDE extensions stopped serving the Gemini Code Assist for individuals, Google AI Pro and Ultra tiers, and Google directs those users to migrate to Antigravity CLI. The practical path today is a Gemini API key of your own, which this calculator prices at about 30% below Claude list rates.
How many hours of agentic coding does a $20 plan cover?
Our working assumption — documented above the comparison table — is that flat subscriptions include roughly 4x their face value in model usage before caps bite, so about $80/mo of list-price tokens for Claude Pro or ChatGPT Plus. At a frontier-heavy mix (about $6.30 per agent-hour) that is roughly 12–13 hours a month, or about 3 hours a week; a fast-and-cheap mix stretches it to roughly 9 hours a week.
Do I need to sign up to use this calculator?
No. It is free, runs entirely in your browser and stores nothing. Move the sliders and the URL updates into a shareable link that replays your exact scenario — send it to a teammate, or copy the ranked summary as markdown straight into a PR or Slack.
Pick the agent first, then price it: try the 30-second agent quiz, or go deeper with the 2026 field guide, OpenCode vs Claude Code and Claude Code vs Cursor.