In This Article
Key Takeaways
- You pay two ways: a monthly seat for a tool (Cursor, Copilot, Claude Code) or per-token for direct API use. Many teams do both.
- Seats cluster low for individuals — Copilot Pro $10, Claude Pro $20, Cursor Standard $40 — and climb fast for heavy agent users (Cursor Premium $120, Claude Max $200).
- API rates vary widely: Claude Sonnet 5 is $2/$10 per million tokens, GPT-5.6 Luna $1/$6, Grok 4.5 $2/$6, while flagship tiers run several times higher.
- The cheapest bill comes from routing: a cheap model for most work, a flagship only when needed, plus caching and batch discounts.
AI coding costs confuse people because two meters run at once. Buy a tool and you usually pay a monthly seat; call a model directly and you pay per token. This guide lays out the current rates for both, dated and sourced, then does the budget math for a solo developer and a small team. Every number below is from a primary source or a recent report, and prices move monthly — model it yourself with our LLM price calculator and API cost estimator before you commit.
Two ways you pay: seats and tokens
A seat is a flat monthly fee for a product — Cursor, Copilot, Claude Code — that bundles some amount of model usage. It is predictable and simple, which is why individuals and most teams start here. A token bill is what you pay when you call a model's API directly from your own code or agent: a rate per million input tokens and a higher rate per million output tokens. Tokens are cheaper at low volume and can be far cheaper at high volume if you route work carefully — but they are variable, so they need monitoring. The interesting wrinkle in 2026 is that seat products increasingly meter heavy usage on top of the flat fee, so the two models are blending.
Seat pricing: Cursor, Copilot, Claude Code, Antigravity
Here are the individual and team seat prices as of July 2026, monthly billing.
Seat pricing, monthly (July 2026)
| Tool | Entry | Mid | Heavy / team |
|---|---|---|---|
| GitHub Copilot | Pro $10 | Pro+ $39 | Max $100; Business $19/seat |
| Claude (incl. Claude Code) | Pro $20 | Max $100 | Max $200; Team $25/seat |
| Cursor | Standard seat $40 | — | Premium seat $120 |
| Google Antigravity | AI Pro $20 | — | AI Ultra $249.99 |
Two changes are worth flagging. Cursor's June 2026 Teams update introduced the $120 Premium seat for heavy agent users (roughly five times a Standard seat's usage) and split usage into first-party and third-party model pools; the new terms hit renewing customers from July 1, 2026. And GitHub moved Copilot to usage-based billing in June, replacing request units with AI Credits at one cent each — base plan prices held, but overages now meter. Claude's own plans (Free, Pro $20, Max $100/$200, Team $25/seat) are on Claude's pricing page. Our Claude Code vs. Cursor cost comparison works a real example end to end.
Token pricing: the API rates
If you build your own agent or call models from code, these are the current published rates per million tokens (input / output). Providers list them; we link each.
API rates per 1M tokens, input / output (July 2026)
| Model | Input | Output | Note |
|---|---|---|---|
| Claude Haiku 4.5 | $1 | $5 | Fast tier |
| Claude Sonnet 5 | $2 | $10 | Intro through Aug 31, then $3 / $15 |
| Claude Opus 4.8 | $5 | $25 | Premium agentic |
| Claude Fable 5 | $10 | $50 | Mythos class |
| GPT-5.6 Luna | $1 | $6 | Fast tier |
| GPT-5.6 Terra | $2.50 | $15 | Balanced |
| GPT-5.6 Sol | $5 | $30 | Flagship |
| Grok 4.5 | $2 | $6 | Configurable effort |
| Llama 4 Maverick | $0.20 | $0.60 | Third-party hosted |
Sources: Anthropic's rates are on the Claude platform pricing docs; OpenAI's GPT-5.6 tiers are on OpenAI; Grok 4.5 pricing is from xAI. Meta wound down its first-party Llama API and now points to third-party hosts, so Llama prices vary by provider — provider comparisons put Llama 4 Maverick around $0.20/$0.60 and Scout lower still.
Budget math for a solo developer
For one developer, seats almost always win on simplicity. A common, capable setup is Copilot Pro at $10 in the editor plus Claude Pro at $20 for Claude Code — $30/month, fully predictable, with generous limits for a single person. If you lean hard on agents all day, one Cursor Standard seat at $40 or a Claude Max plan at $100 removes most limit anxiety. Only reach for the raw API if you are running scripted or batch workloads where token math beats a flat seat. Model your specific mix in the AI cost calculator before upgrading — most solo developers overbuy.
Budget math for a team
Teams should separate two budgets: seats for interactive coding, and tokens for any automated pipeline. For a ten-developer team, interactive seats might be ten Copilot Business at $19 ($190/month) or a mix of Cursor Standard and Premium seats depending on who runs heavy agents — that is where the $40-vs-$120 split matters, since a handful of Premium seats can cost more than everyone else combined. Separately, any CI, review, or batch automation should call the API directly and route by difficulty. A single guardrail saves the most money here: cap per-seat overages and alert on token spend, because usage-based billing turns a runaway agent into a runaway invoice.
Five levers that cut the bill
- Route by difficulty. Send routine work to a fast tier (Haiku 4.5, Luna) and reserve flagships for genuinely hard tasks. Our model-tier guide shows how.
- Cache prompts. Anthropic's prompt caching can cut cache-hit input cost by up to 90% — see how prompt caching works.
- Batch non-urgent jobs. Batch APIs commonly halve token cost for work that can wait.
- Use open weights for volume. A self- or plan-hosted open-weight model can run high-volume coding at a fraction of flagship cost; see our open-weight vs. frontier API guide.
- Cap and alert. Set overage limits and spend alerts on every metered plan before you scale usage.
Predictable beats cheap until you measure
A flat seat you understand is worth more than a token bill you cannot predict — right up until you actually measure your usage. Start on seats, watch your real consumption for a month, and only move workloads to the API where the math clearly wins. The teams that overspend are the ones that guessed; the teams that save are the ones that instrumented their spend and routed accordingly.
Model your real bill in two minutes
Plug in your monthly tokens and compare providers side by side — free, no signup.
Open the LLM Price CalculatorSources: Cursor Teams pricing; GitHub Copilot billing; Claude pricing and API rates; OpenAI GPT-5.6; xAI Grok 4.5; Llama provider pricing. All prices are July 2026 and change frequently; verify at the source.
Common questions
Is it cheaper to buy a seat or use the API? For one person coding interactively, a seat is almost always cheaper and simpler. For automated or high-volume workloads, direct API access with careful model routing usually wins. Many teams run both — seats for humans, tokens for pipelines.
Why did my Cursor or Copilot bill change in 2026? Both moved toward usage-based components. Cursor added a $120 Premium seat and split usage pools in its June 2026 update; GitHub replaced Copilot request units with metered AI Credits. Base prices largely held, but heavy usage now meters on top.
What is the single biggest way to cut cost? Route work by difficulty. Output tokens on a flagship run roughly five times a fast tier, and most coding tasks do not need the flagship. Add prompt caching and batch discounts and the savings compound.