Claude takes the most interesting approach to its free tier of any major AI tool. It doesn't give you the strongest models, but it packs nearly every tool into Free. You can code, search the web, connect external services, and run scripts—just not with Opus, and not with Claude Code.
This article is based on Anthropic's official pricing page: what Claude Free can actually do, where its limits are, and when upgrading makes sense. Prices below are list prices as of 2026-08-12. Check the Claude website for the latest rates.
Model access: Haiku + Sonnet are enough; Opus stays behind the paywall
Claude Free gives you every Haiku and Sonnet version across all features with full 200K context—covering most coding and document work, but the entire Opus lineup and Fable 5 require paid plans. Per Claude's official pricing page, Free tier model access is straightforward (source):
| Model | Free Access? | Notes |
|---|---|---|
| Haiku 4.5 | ✅ | Fastest, lightest—great for high-frequency tasks |
| Sonnet 5 / 4.6 / 4.5 / 4.0 | ✅ | Strong coding, your daily driver |
| Opus 5 / 4.8 / 4.7 / 4.6 / 4.5 | ❌ | Requires Pro or above |
| Fable 5 | ❌ | Requires Team/Enterprise |
All Haiku and Sonnet versions work across every feature—chat, code, file analysis—no gating behind specific feature entrances. The full 200K context window is available on Free, handling very large code files and long documents.
What this means in practice: if you use Claude mostly for writing code, fixing bugs, and code review, Sonnet covers everything. Opus's real value shows up in multi-step reasoning, large-scale architecture refactoring, and complex agent orchestration—tasks where Sonnet clearly struggles.
Tool ecosystem: the most complete free-tier tool chain
Claude Free has the most complete tool ecosystem of the three major AI tools—web search, code execution, MCP connectors, desktop extensions, Extended Thinking, and six other features all available without restrictions—Anthropic's strategy is to gate models, not tools. This is Claude Free's biggest strength. Anthropic's strategy isn't to gate features but to let you experience the full product and upgrade because you want Opus (source).
Available on Free, no restrictions:
- Web search: real-time information retrieval
- Code execution: sandboxed Python environment for creating files and running code
- MCP connectors: integrate Slack, Google Workspace, and external services via remote MCP
- Desktop extensions: full macOS and Windows desktop app functionality
- Extended Thinking: complex reasoning for harder problems
- Cross-conversation memory: Memory feature for persistent preferences
- Projects: organize chats and documents
- Artifacts: interactive code previews
- Voice mode: voice conversations
- Incognito chats: private mode
Compared to ChatGPT Free locking GPT-5.6 Terra behind desktop Work and Codex, or Cursor Hobby's agent and tab limits—Claude Free is the most open of the three. You get the complete Claude experience; the only thing missing is the top-tier model.
Usage limits: where the "twice full" ceiling actually sits
Claude doesn't publish exact numbers, using a dual-layer '5-hour rolling window + weekly cap' system—Free gets roughly 50% of paid limits: Haiku stretches the quota far, Sonnet hits the ceiling faster, and Extended Thinking burns through it fastest. Claude doesn't publish exact numbers, using a dual-layer "5-hour rolling window + weekly cap" system (source). Free gets roughly 50% of paid-tier limits.
What that feels like:
- Mainly using Haiku: 50% quota goes surprisingly far. Haiku is fast with low per-request cost—high-frequency document work, questions, and simple code edits stay under the ceiling.
- Mainly using Sonnet: medium tasks (editing a few files, analyzing a hundred lines) at a dozen per day will trigger the 5-hour window. Sonnet costs noticeably more than Haiku but delivers noticeably better code quality.
- Heavy Extended Thinking: burns through the window rapidly. This feature is designed to be resource-intensive.
What happens when limits hit: no overage option. You wait for the window to reset (gradually releasing capacity every 5 hours). Getting interrupted mid-workflow is the worst experience on Free—and the main reason people upgrade.
What Pro adds: three things Free doesn't have
Pro ($20/mo) adds three things Free doesn't have: the full Opus lineup, Claude Code/Cowork/Design applications, and roughly 2x usage limits—the upgrade signal isn't hitting the cap, it's when you're compromising code quality because you can't use Opus. Pro at $20/mo (or $200/year) adds three key capabilities (source):
- Full Opus lineup: Opus 5/4.8/4.7/4.6/4.5 all unlocked. For large codebase refactoring, extended reasoning, and complex agent orchestration, Opus does things Sonnet simply can't.
- Claude Code / Cowork / Design: Claude Code is Anthropic's AI coding agent, Cowork is a collaboration workspace, and Design is a design tool. These three applications are completely unavailable on Free.
- More usage: roughly 2x Free's limit, plus priority access during peak times.
The upgrade litmus test isn't "did I hit the limit"—it's "am I compromising code quality because I can't use Opus." If Sonnet handles your daily needs, Free is the best version for you. When you hit tasks Sonnet clearly struggles with, Pro becomes worth it.
The right approach: use Claude Free with Sonnet as your daily driver. When you encounter tasks too large or complex for Sonnet to handle well, that's your signal to upgrade. Claude Free has the lowest barrier and closest-to-paid experience of the three major free tiers.