Is Claude Code free? The installer is free, but Claude Free doesn't include the coding agent. I make Octomind, which offers a local-model route of its own, so weigh my framing with that in mind. I checked Anthropic's current plan and API documentation on October 27, 2026, and rechecked the plan lineup November 3, 2026.
The short answer: Claude Code pricing for Claude models starts with Pro at $20 monthly ($17 a month billed annually); move to Max at $100, or Ultra at $200, if shared usage limits stop your work. Team and Enterprise seats include Claude Code, while Console, Bedrock, Google Cloud, and Microsoft Foundry bill by usage. A local model can make the Claude Code harness cost nothing per token, but it is not Claude. The cheapest route depends on how many tokens you use and whether you need Anthropic's models.
Is Claude Code free to install or use?
You don't need to pay to install the CLI; you do need a Claude subscription, Console account, or supported cloud-provider access to use it (Anthropic quickstart, 2026). The free Claude account includes chat, but Anthropic's plan table marks Claude Code “No” for Free and “Yes” for Pro, Max, and Ultra (Anthropic pricing, 2026).
A new Claude Console account also gets a small amount of free API credit to test the API; Anthropic doesn't publish a standard amount (Claude Platform pricing, 2026). Treat that as a trial, not a free tier.
Anthropic doesn't list a standard Pro or Max student discount; university-sponsored Education accounts vary by school, and Anthropic's guide doesn't promise Claude Code access specifically (Pro plan help, 2026, Education help, 2026).
Claude Code pricing: which plans include it, and what do they cost?
Pro is the lowest-priced of the Claude Code plans; the Claude Max price starts at $100 a month. The plan usage pool is shared with Claude chat, so a long chat can use capacity you expected to have for the terminal (Anthropic pricing, 2026).
Prices below are US list prices checked October 27, 2026; regional prices and taxes can differ. Anthropic can change them.
| Route | Price | Claude Code access | What the price buys |
|---|---|---|---|
| Claude Free | $0 | No | Claude chat, not Claude Code |
| Pro | $20/month or $200/year ($17/month effective) | Yes | Shared five-hour and weekly plan usage |
| Max | $100/month | Yes | 5x Pro usage every 5 hours and every week |
| Ultra | $200/month | Yes | 10x Pro usage every week; no 5-hour limit |
| Team standard / premium | $25 / $125 per seat monthly; $20 / $100 annually | Yes | Team seats; premium has five times standard-seat usage |
| Enterprise | $20 per seat/month, annual, 20-seat minimum, plus usage | Yes | Seat fee plus API-rate usage |
| Claude Console API | Per token; no subscription | Yes, with API billing | Usage billed to your Console account |
| Bedrock / Google Cloud / Microsoft Foundry | Per token on the cloud account | Yes, through provider setup | Usage billed by the cloud provider |
Sources: Anthropic pricing, 2026, Max plan details, 2026, Anthropic deployment options, 2026.
Anthropic says paid plans have a rolling five-hour limit and a weekly limit. There's no fixed number of messages: model choice, context length, and task complexity change how quickly you use the shared pool (Anthropic pricing, 2026).
Max's monthly API credits are separate: Anthropic says they can't be used for Claude Code (Pro and Max billing help, 2026).
Claude subscription vs API: is API billing cheaper than Pro or Max?
It can be, if your monthly use is light. Sonnet 5.5 bills input at $2 per million tokens and output at $10 per million, so a month of API use costs:
$2 × input millions + $10 × output millions
These are scenarios, not measured developer profiles. They assume all input is billed at the uncached rate, use one model, and exclude taxes or other cloud fees. Prices were checked October 27, 2026 (Anthropic API pricing, 2026).
| Example month | Input + output tokens | Sonnet 5.5 API estimate | Price comparison |
|---|---|---|---|
| Light | 1M input + 0.1M output | $3 | Below Pro's $20 monthly price |
| Daily | 10M input + 1M output | $30 | Above Pro; below Max and Ultra |
| Heavy | 50M input + 5M output | $150 | Above Max; below Ultra |
At these Sonnet rates, API spend matches $20 when input millions + 5 × output millions = 10; it matches $100 at 50 and $200 at 100. That compares dollars only. A subscription buys a limited shared usage pool; API billing doesn't give you the same capacity or predictable cap.
Claude Code's prompt cache can lower the input side substantially. Sonnet 5.5 cache reads cost $0.10 per million tokens, while a five-minute cache write costs $2.50 per million (Anthropic API pricing, 2026). The example deliberately leaves cache reads out, so check your actual /usage breakdown before using it as a forecast.
Anthropic reports $150–$250 per developer per month on average across enterprise deployments, but that's not a prediction for an individual user (Claude Code cost guide, 2026).
Can I run Claude Code for free with a local model?
Yes, if by “Claude Code” you mean the coding agent interface rather than Claude models. Ollama documents a local-model integration with Claude Code; its January 2026 setup requires Ollama v0.15 or later and uses ollama launch claude to pick and start a model (Ollama, 2026). This is a docs-only setup in this article, not a test I ran.
The model runs on your hardware, so there's no provider token bill, but hardware, electricity, setup time, and model quality still have a cost. Ollama recommends at least 64K context for coding tools; pick a local model that fits your machine.
The agent interface remains Claude Code, but the reasoning model is an open model, not a Claude model (Ollama, 2026). Anthropic's gateway docs say Claude Code “doesn't support routing Claude Code to non-Claude models through any gateway”, so treat this as an Ollama-documented compatibility route without Anthropic support (Anthropic gateway docs, 2026).
For a free coding agent that can use local Ollama models, Octomind is another option: it's Apache 2.0 software, and its documented command is octomind run developer:general -m local:<model> (our local Ollama setup). The setup also requires pointing supervisor and compression calls at the local model if you want all inference to stay on the machine.
If you don't have local hardware, Octomind Hub is pay-as-you-go: signing up, keys, and the CLI cost nothing, model calls bill at provider pricing from prepaid credits, credit top-ups add 3%, and the key pauses at a $0 balance (free Hub details). It's a separate coding agent route, not Claude Code access.
OpenCode is a fair alternative if you want a different open-source agent: its docs list 75+ providers and local model support (OpenCode provider docs, 2026). The tool may cost nothing, but hosted models still bill through their provider.
Older Gemini CLI guides are stale as a free consumer route: Google ended “Login with Google” access for individuals on June 18, 2026 (Google for Developers, 2026); API-key access is a separate route.
What happens when I hit a limit or an API bill grows?
On Pro or Max, check /status for your included plan usage. When you reach a five-hour or weekly limit, wait for reset, move up a plan, or enable usage credits; those credits bill at API rates and have their own monthly spend limit (Pro and Max billing help, 2026, usage credits help, 2026). If you don't want extra charges, decline the API-credit option.
For API billing, set a monthly spend limit in the Console before a long run (Anthropic API spend limits, 2026). /usage estimates the current session cost; the Console Cost page (“Manage costs for your organization”) is authoritative for API users (Claude Code cost guide, 2026).
Check your login if the numbers look wrong: a stray ANTHROPIC_API_KEY overrides a subscription login and causes API usage charges (Pro and Max billing help, 2026).
On Claude Pro vs Max: pick Pro if you want Claude models and your work fits the shared plan limits. Pick Max if you keep hitting those limits and the additional capacity is worth the fixed monthly price, and Ultra if Max still runs out; Anthropic's own advice to Max users who hit limits is to consider Ultra. Pick Console or cloud API billing if usage is light or spiky and you want to pay only for tokens, with a spend cap. Pick a local model if zero model-provider spend and keeping code on your machine matter more than Claude model quality.
For a closer look at plan limits, see our Claude Code rate-limit guide; if your main concern is repeated context charges, see how prompt caching changes token costs.
