AI coding assistants advance as Anthropic merges 46% of Claude Code's pull requests, DeepSeek launches open source rival

AI coding assistants are advancing rapidly as Anthropic and DeepSeek expand their agent capabilities. Anthropic successfully tested Claude Code on its own software, merging 46% of 388 AI-generated pull requests. Meanwhile, DeepSeek launched DeepSeek Harness v0.1, an open-source, modular rival to Claude Code, alongside the official release of its agent-focused DeepSeek-V4-Pro model. Concurrently, DeepSeek is restructuring its API pricing on August 16, 2026, introducing peak and off-peak rates that increase token costs by 50% to over 1,100%.

DeepSeek Harness open-source launch

• ▪DeepSeek Harness is built on Cordis, a framework designed around composable plugins, allowing developers to swap out models, tools, sandboxes, filesystems, and user interfaces.
• ▪DeepSeek launched DeepSeek Harness v0.1, an open-source agent harness under the MIT license, as an alternative to integrated coding-agent environments like Anthropic's Claude Code.
• ▪DeepSeek Harness can inspect repositories, edit files, execute shell commands, search files and the web, maintain plans, invoke skills, delegate work to subagents, and enforce approval policies.

DeepSeek V4-Pro agent capabilities

• ▪DeepSeek-V4-Pro-0813 features native support for the OpenAI Responses API, integration with Codex, and three configurable reasoning effort levels: Non-think, Think High, and Think Max.
• ▪DeepSeek reported that DeepSeek-V4-Pro-0813 scored 87.9 on Terminal Bench 2.1, 74.1 on Toolathlon-Verified, 71.1 on DSBench-FullStack, and 67.2 on DSBench-Hard.
• ▪DeepSeek released the official version of DeepSeek-V4-Pro-0813, transitioning the 1.6-trillion-parameter model from preview to general availability with enhanced agent capabilities.

Claude Code autonomous maintenance experiment

• ▪Claude Code executed twelve specialized maintenance routines during the Anthropic experiment, including a Crash Fuzzer to trigger and fix app crashes, a Dup Unifier to merge similar abstractions, and a Dead-Code Remover.
• ▪During the maintenance experiment, Claude Code created 388 pull requests across Anthropic's repositories, and 180 of those pull requests were merged after human review, representing a 46 percent merge rate.
• ▪Anthropic engineer Boris Cherny conducted an experiment where Claude Code ran daily maintenance routines on Anthropic's in-house applications via a dedicated Slack channel.

AI coding agent competitive landscape

• ▪DeepSeek Harness and Anthropic's Claude Code represent a competitive shift where AI providers compete over the agent execution layer that controls how models manipulate files and execute workflows.
• ▪DeepSeek Harness emphasizes modularity where the model is a swappable plugin, whereas Claude Code and OpenAI's Codex offer more vertically integrated developer experiences.

DeepSeek API pricing restructuring

• ▪DeepSeek is abandoning its flat API pricing on August 16, 2026, to introduce peak and off-peak rates, resulting in price increases ranging from 50% to over 1,100% depending on the model and token category.
• ▪Under the new pricing structure, DeepSeek-V4-Pro cache-miss input tokens will rise from $0.435 per million to $0.66 off-peak and $1.32 during peak hours, while output tokens will rise from $0.87 per million to $1.98 off-peak and $3.96 at peak.

2 sources