Amp vs Claude Code: pricing, models and fit

We make Shuriken. We tried to be fair; corrections welcome at office@shuriken.sh.

Quick answer#

Amp vs Claude Code is mostly a question of how you want to pay. Amp is a multi-model coding agent with a free Hobby tier: you bring your own ChatGPT or Claude subscription or API keys, or buy credits that Amp bills at the providers' API prices with no markup. Claude Code is Anthropic's own agent, for Claude models only, included in Pro ($20 a month) and Max ($100 or $200) with a 5-hour and a weekly limit.

Pick Amp if you want an agent that picks among GPT, Claude and other models for you, and a bill that follows what you use. Pick Claude Code if Claude is the model you want and you'd rather pay one flat price for it.

At a glance#

Read from Amp's and Anthropic's own pages on 24 October 2026. US dollars a month, as listed; Anthropic's prices are before tax, and Amp's pricing page doesn't say.

There are no benchmarks on this page. Amp changes the model behind each mode as new ones come out, so a score from one month says little about the next.

What Amp is in October 2026#

Amp's introduction lists three places to work. The web app at ampcode.com needs no install and runs each thread in an orb. The Mac, iPhone and iPad apps are in beta, need version 26 of macOS or iOS (and Apple silicon on a Mac), and add notifications and dictation (Amp's app docs). The CLI runs on macOS, Linux and Windows through WSL, and installs with curl -fsSL https://ampcode.com/install.sh | bash (Amp CLI docs).

Amp describes orbs as "Machines that Amp creates per thread, with your code, tools, and plugins already set up", and adds: "Send a prompt, close your laptop, and the agent keeps working." From the CLI, amp --executor local, orb or runner:<id> chooses where a thread runs. Runners on your own machine and self-hosted orbs are free.

In an editor, Amp works through its CLI, which connects to VS Code, Cursor, Windsurf, Zed and Neovim: run amp, press Ctrl+O and choose ide connect, and Amp sees the open file and edits through the editor with undo.

Modes: Amp picks the model, Claude Code lets you#

Amp asks you how hard the task is, not which model to use. The Dial has four modes. Low is for "small, well-defined tasks", medium is the default, high "spends more time reasoning", and ultra is for "hard, open-ended work where the outcome is clear but the path is not". Each mode bundles a model, a reasoning effort, a system prompt and tools, and Amp changes the model behind a mode as better ones come out (Modes & Models).

A thread keeps the mode it started with. Amp's docs give two reasons: most models "get confused when they continue a conversation that another model started", and a switch would invalidate the prompt cache. To change mode you start a new thread. You can still pin a model and effort per mode under Settings › Mode Dial › Tune Modes.

Claude Code works the other way round. You pick a model with an alias such as opus, sonnet, haiku or fable (model configuration), and /model switches it during a session. Every choice is a Claude model. Through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry it is still Claude.

Amp pricing in 2026#

Amp's pricing page has four tiers, read on 24 October 2026:

Amp's pricing docs say "Paid tiers include agent usage, orb usage, and every Amp product feature". Included usage resets each month and doesn't roll over. Past it, you buy credits, and those need no paid tier at all. "For individuals and non-enterprise workspaces, Amp does not add a markup to providers' API prices." Amp's own example: a thread that uses $2 of Anthropic API usage and $0.50 of OpenAI usage takes $2.50 from your credits. Web search and other non-model tools use credits too, and purchased credits expire after twelve months.

Is there still an Amp free tier?#

Hobby is free, but it isn't free model use: you pay your own subscriptions, keys or credits. The free daily grant is a separate thing. On 30 March 2026 Amp dropped the ads from Amp Free and said most people with "the $10 daily free usage of Amp" would keep it. An update dated 23 May 2026 says Amp is "pausing or reducing further Amp Free's daily allowance for more users". The current pricing page doesn't list a daily grant, so check your own account before you count on one.

Claude Code's plans and limits#

Claude Code is in every paid Claude plan and not in Free. Pro is $20 a month, or $17 billed yearly; Max is $100 for 5x Pro's usage or $200 for 20x (Claude pricing, Max plan page). Anthropic's help page says "Both Pro and Max plans have a five-hour session limit and a weekly limit", shared with Claude chat, and Max adds a separate weekly limit for Fable. At a limit you can turn on usage credits, move up a plan, wait for the reset, or switch to a Console account and pay standard API rates.

Max also comes with $100 or $200 a month in API credits, but the Claude Code help page says those credits "can't be used in Claude Code or the Claude apps". Claude Code on Max runs on the Max limits, not on those credits.

Claude Code's cloud sessions run in Anthropic's machines and keep going after you close the laptop, like orbs. The difference is the bill. A cloud session "counts toward your plan's usage limits" with "no separate charge for the cloud VM" (cloud docs); an Amp orb is billed by the minute, from a paid tier's orb time or from credits.

Whether a flat plan or an API key costs you less depends on your own week of work; our API key vs subscription guide prices one week of your usage both ways.

Can you use a Claude subscription in Amp?#

Amp says you can, in one mode. Its announcement of 10 October 2026 adds a Claude Code mode that "uses the Claude Agent SDK instead of Amp's own agent". You link a Claude subscription in Settings › Model Routing, and orbs and runners use it. Amp's Dial docs draw the line: "Amp's own modes, such as medium, high, and ultra, and any raw models are never covered by your Claude plan. Only the Claude Code mode uses your Claude plan." Orb minutes are not covered by the Claude plan either.

Anthropic has published two statements that bear on this, and we quote them without ruling on Amp's setup. Its Claude Code legal page says: "Anthropic does not permit third-party developers to offer Claude.ai login into their own applications, or to route requests through Free, Pro, or Max plan credentials on behalf of their users." Its help article on the Agent SDK, in an update dated 7 October 2026, says: "You can still use the Claude Agent SDK, claude -p, and third-party apps with your subscription limits." The legal page adds that Anthropic "reserves the right to take measures to enforce these restrictions and may do so without prior notice", and it sends questions about permitted sign-in methods to Anthropic's sales team. If your plan matters to you, ask Anthropic before you link it. How other tools handle a Claude sign-in is laid out in using your Claude plan in other tools.

ChatGPT is simpler on Amp's side: sign in with ChatGPT and allow token sharing, and OpenAI model use bills to your ChatGPT plan. A one-click preset pins low, medium and high to OpenAI models. Ultra keeps Amp's own routing.

Threads and sharing#

Every Amp thread, a local CLI one included, gets a URL at ampcode.com and opens in the web app, the CLI and the Mac and iOS apps (Amp threads docs). Share sets it to private, workspace, group (Enterprise only) or unlisted, which means anyone with the link. Without a workspace, new threads are private. Adding .md to a thread's URL reads it as Markdown.

In Claude Code, the cloud sessions can be shared: on Pro and Max a session is Private or Public, where Public means any signed-in claude.ai user with the link, and on Team and Enterprise it is Private or Team. Anthropic warns that a session "may contain code and credentials from private GitHub repositories", so read it before you share it.

Which one fits whom#

Amp fits if you'd rather choose how hard a task is and let the agent pick the model, if you already pay for ChatGPT and want to use that plan in Amp, if you want to pick, thread by thread, whether the agent runs on your laptop, in an orb or on your own runner, or if your team reads each other's threads. It also fits light or uneven use: Hobby costs nothing, and credits bill what a thread used, at the providers' price.

Claude Code fits if Claude is the model you want and you code most days. A flat $20 to $200 plan with 5-hour and weekly limits is easier to budget than a per-thread bill, and the cloud sessions come with no machine charge. The wider field, both of these included, is in Claude Code alternatives, and the plans across providers are priced in best AI coding subscription.

Running Amp and Claude Code together#

Both read the same instruction file if you let them. Amp loads AGENTS.md, and falls back to AGENT.md or CLAUDE.md in a folder that has no AGENTS.md (Amp's AGENTS.md docs). Claude Code reads CLAUDE.md, and since v2.1.277 an AGENTS.md when there is no CLAUDE.md (memory docs). Keep one AGENTS.md and both agents start from the same instructions.

Shuriken is a harness that runs coding agents on your own computers and servers, and Amp and Claude Code are both in its agent catalogue (the full list). Claude Code runs as a built-in Shuriken chat. Amp opens as a Shuriken chat through the Agent Client Protocol, by way of the amp-acp adapter, which has to be installed on that computer next to the Amp CLI. It gets the same queue and phone view. Every chat except a project's Main chat gets its own branch in its own git worktree, so an Amp chat and a Claude Code chat can take the same task side by side without editing each other's files. Each agent runs on your own sign-ins and keys: Claude Code on your Claude plan, Amp on an Amp API key you add as a source. Shuriken never bills model usage.

If you want to try both on your own code, download Shuriken and give the same prompt to an Amp chat and a Claude Code chat.