Rakuten Group
Rakuten reports feature time to market cut from 24 working days to 5 with Claude Code
Rakuten, the Japanese internet services group, rolled out Anthropic's Claude Code across its engineering teams. Its own blog reports that average time to market for new features fell from 24 working days to 5 days.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
From 24 working days to 5 days
Average time to market for new features, before and after (Rakuten-reported)1
99.9%
Numerical accuracy of that implementation versus the reference method1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Rakuten runs more than 70 businesses, from e-commerce to travel, fintech and communications, with thousands of developers serving millions of customers. It wanted new features to reach customers faster without lowering quality or security. Its leaders say speed and return on investment are the key measures.
What they built
Engineering teams use Claude Code, Anthropic's coding agent, across the development cycle: writing unit tests, mocking interfaces, building components, fixing bugs and writing documentation. New team members also use it to understand large codebases. It also provides AI-powered code review on pull requests, and engineers run several sessions at once.
How they did it
Rakuten tested the tool on a hard task. A machine learning engineer asked it to implement a specific method inside vLLM, a very large open-source library with 12.5 million lines of code. Claude Code worked for seven hours in a single run with only occasional guidance. Rakuten says the result matched the reference method with 99.9% numerical accuracy. After that, the company redesigned its development workflows around the tool and, following pilots, committed to widening access. It also had non-engineers try it, with coding guidelines acting as a guardrail.
Results
Rakuten reports that the average time to market for new features fell from 24 working days to 5 days, which it calls a 79% reduction. The same post reports the seven-hour autonomous run and the 99.9% accuracy result. Its engineers describe tests being generated instantly and several tasks running in parallel.
Lessons for your project
Test on a hard, real task first. Rakuten used a seven-hour run on a large open-source codebase as its proof point before committing to wider access.
Redesign the workflow, not only the tool. Rakuten says it rebuilt development around Claude Code instead of adding it to old processes.
Run several sessions in parallel. Engineers hand some tasks to Claude Code while they work on another, which is where Rakuten sees capacity gains.
Choose a business measure, not a coding measure. Rakuten tracks time to market because it wants to judge AI by how fast value reaches customers.
Caveats
- All figures come from Rakuten's own blog post, which is co-presented with Anthropic. No independent source confirms them, and the post does not say how time to market was measured, over which teams or period.
- The 99.9% figure is one engineering test (a single task in the vLLM library), not a measure of the whole rollout.
- The post says leadership is committed to scaling Claude Code after successful pilots, so the scale of use at publication is not stated as a number.
- No go-live date is given. The year 2025 is the year of the blog post (June 26, 2025). The model is named only as Claude Code with no version.
Sources
- Rakuten accelerates development with Claude Code
Rakuten Today (Rakuten Group) · 2025-06-26 · Company blog · rakuten.today
- Rakuten customer story
Anthropic · Undated · Vendor case study · claude.com
Summarized in our own words from the sources above. Spotted an error? See corrections.