Anthropic Cleans Up Claude Code: The Fix That Restored Performance (And Reset Your Quotas)
After a period of frustration among developers who experienced a sharp drop in work quality with Claude Code, Anthropic has confirmed the issues and released a new version. Is the promising Agentic Coding model back on track? Here is everything you need to know.
In recent weeks, the developer community using Claude Code – Anthropic's revolutionary CLI tool – began to feel that something was not working right. What was supposed to be an autonomous AI agent with deep code understanding started to exhibit slowness, make mistakes on simple tasks, and sometimes simply "freeze" in the middle of complex processes.
Many wondered if the model itself (Claude 3.5 Sonnet) had undergone degradation, but it now turns out that the story is quite different. Anthropic officially confirmed that the problem was not in the AI's "brain," but in the infrastructure supporting it.
It is not the model, it is the pipeline: What really happened?
The complaints that accumulated in forums and on GitHub pointed to three major issues in Claude Code and the Agent SDK (the kit used for developing agents, upon which the Cowork tool is also based). These issues caused poor context management, which led to the agent losing its way in large projects.
In the newly released version, 2.1.116, Anthropic addressed these flaws at the root. It is important to understand: when working with an AI agent in the terminal, performance depends heavily on how the SDK interprets system outputs and feeds them to the model. A small bug in the SDK can make the smartest model in the world look confused, and that is exactly what happened here.
What is new in version 2.1.116?
The current update is not just a bug fix, but a significant improvement in user experience:
- •67% improvement in session resume speed: Developers working on log files or large sessions (over 40MB) will feel a significant difference in the response speed of the
command./resume - •Transition to native binary: Starting with the latest versions, the CLI no longer runs just as wrapped JavaScript code, but as a dedicated binary file for each operating system (Windows, macOS, Linux), which reduces tool startup time and memory consumption.
- •"Thinking" status reflection: An intuitive progress indicator has been added, showing what Claude is doing while it is "thinking," instead of leaving the user facing a blinking cursor.
- •Fix in GitHub Integration: The handling of GitHub API rate limits has been improved, preventing the agent from entering loops of failed retry attempts.
The compensation: Full quota reset
In an unusual move, Anthropic decided to compensate subscribers who paid for a service that did not deliver in the recent period. All subscribers received a full reset of usage limits. This is a significant step, as Pro and Max subscribers are limited by weekly and daily quotas, and such a reset allows "burning" tokens on complex projects without fear, at least for the coming days.
How to proceed from here?
If you use Claude Code as part of your development routine, the first action you should take is to update. Simply run in the terminal:
claude updateor perform a reinstall via npm to ensure you are on version 2.1.116 or higher.
The bottom line: This case is an excellent reminder that in the world of Agentic AI, the quality of the model is only half the equation. The infrastructure, the SDK, and the way the tool communicates with your file system are just as important. Anthropic has demonstrated maturity here – both in quickly identifying the technical issue and in the service-oriented step of resetting the quotas.
Relevant links: