Anthropic
Does not work from Russia
Terminal
Claude Code
The most mature subagent system; included in any paid Claude plan.
Verified: Sep 29last change: Sep 28next price check: Dec 28
Price
Max $100–200 per seat
License
closed-source
Model
Opus 5.5 · Fable 5.1 · Sonnet 5.5
model under the hood
–
Facts with sources
- Pro at $20 is enough to try it out: on a real task the limits run out within a couple of hours, so plan the team budget from Max. Source (in Russian)
- The subscription gives you only Claude: you can’t switch models within it. Source (in Russian)
- On July 21 the system prompt was cut by about 80%: newer models handle context better than a list of rules. Source (in Russian)
- Anthropic asks for identity verification with a document and a selfie; corporate signs of use from Russia lead to loss of access. Source (in Russian)
Strengths
- In Tech With Tim's tests Opus 5 went deeper: 27 tests for a habit tracker and its own proof tool when splitting a 5,040-line file.
- Included in every paid Claude plan; runs in the terminal, VS Code, JetBrains, the web, a mobile app, Slack and GitHub Actions.
- CLAUDE.md memory, subagents, hooks, MCP, background agents and cloud-scheduled routines, per Anthropic's docs.
- The Claude Code team says it now does 70 to 80% of its work through the Slack agent and daily routines that triage feedback.
Weaknesses
- The $20 Pro plan gives only Sonnet 5; Opus 5 and Fable start at the $100 Max plan (Tech With Tim).
- Splitting a 5,040-line file took Opus 5 36 minutes, almost 5x longer than Codex on the same prompt.
- Usage is shared with the Claude chat app: about 45 prompts per 5 hours on Pro by community estimate, Anthropic publishes no numbers.
- Jordan Urbs cancelled the $200 Max plan over vendor lock-in: two of his 700M-token days would have cost about $1,800 at API prices.
Example from practice
Split a 5,040-line file without breaking anything
Tech With Tim gave Codex and Claude Code the same prompt: split a 5,040-line file from the Textual library and keep everything working. Opus 5 got it done but took 36 minutes, almost 5x longer than Codex, because it also wrote a proof tool showing that behaviour had not changed. Haiku did not finish: it created the new files but never removed the original. The reviewer's verdict is that you pay for that depth in time and subscription quota.
source: Tech With Tim, "Codex vs Claude – an Honest Comparison"
What reviewers say
Reviewer opinions on their own tasks and hardware, not measurements. Links go to the videos; transcripts live in docs/raw-data.
- Codex vs Claude – an Honest Comparison · Tech With Tim, Aug 27The reviewer's view: pick Claude Code for depth, since Opus 5 did more than asked on all five tasks, but expect it to be several times slower and heavier on quota.
- How the Claude Code team uses Claude Code · Claude, Sep 2Anthropic's own engineers describe moving from watching every tool call to handing goals to a Slack agent and cloud routines; this is the vendor talking about itself.
- Why I Cancelled My Claude Code Subscription · Jordan Urbs, Sep 6The author thinks the $200 Max plan is too good a deal to last and moved to OpenCode with open-weight models so his workflow no longer depends on Anthropic.
- Is Pi the Best Coding Agent? Pi vs OpenCode vs Claude Code · pookie, May 24The author left Claude Code for the minimal Pi, citing a 14,000-token system prompt and subagents whose work you cannot see.
- How auto mode works with Claude Code · Claude, Aug 4The vendor explains auto mode: a separate classifier judges risky actions, deny/ask/allow rules run before it, and admins set managed settings developers cannot remove.
- How Ramp engineers work with AI agents at every step · Claude, Aug 6Ramp (with the vendor): a fixed set of hard tasks rerun on every model, read-only keys for agents, no per-person budgets; the claimed CI drop from 18 to 6 minutes is unverified.
- DoorDash gave every employee Claude Code · Claude, Jul 7DoorDash (with the vendor): every manager had to ship production code with an agent, a fast procurement path, written write-ups of failures; the bottleneck moved to product reviews.
- How Spotify runs agents across 20M+ lines of code · Claude, Jun 29Spotify (with the vendor): an LLM judge lifted automated PR success from 20 to 30 percent to 80, then was removed; auto-merge forced investment in test automation.
Who it is for
- Engineer who knows the repo, branches, and terminal
- Team where Anthropic is available and wants one tool for everyone
Models in this agent
Score is from our manager-task benchmark; models without a score are not measured yet.
What changed
- BugHunt: Sonnet 5.5 at max effort fixed the most bugs – 55.5 of 105 on average against 41.7 for Opus 5.5 – but a run cost about $135 against $59 and took almost four hours against 67 minutes: for every fix it writes a test and confirms the test fails with the fix reverted →
- Claude Opus 5.5 released: $4 per million input and $20 per million output tokens – 20% cheaper than Opus 5, with cache reads 60% cheaper →
- Redesigned Projects in Claude Code entered beta for select Pro and Max subscribers on cloud sessions; others can join the waitlist →
- Anthropic reworked Claude pricing: Pro $17 a month billed yearly, Max from $100, Team $20 / $100 per seat, Enterprise $20 per seat plus API usage →
- Claude Code v2.1.269: a configurable limit of up to 256 concurrent agents per Workflow run; claude plugin eval runs plugin evaluations with scores and reports →
- Claude Code gained an Advisor mode: a second Opus checks the first one's work, the request context grew to 1M tokens →
- Anthropic released Claude Fable 5.1: available in Claude Code as the top-tier model at $10 / $50 per million tokens
- Claude Code switched to auto mode by default: a classifier judges commands instead of allow/deny prompts →
- Anthropic released the Claude apps gateway for AWS: team access to Claude Code and Desktop via SSO, spend limits and policies →
- Claude Code auto mode explained: a separate classifier judges risky actions, admin rules cannot be removed →
- Claude Code now defaults to Opus 5; Fable 5 and Sonnet 5 are switched manually
- The Claude Code system prompt was cut by about 80% for frontier models
How to brief an agent like this
The four-line brief is the Foundation module: prompt structure, LLM agents and MCP, an AI policy for the team. It starts with the free open module.