Tag: #coding-agent22 results
Anthropic has released the Claude Fable 5 coding model, achieving a record-breaking 95.0% on the SWE-bench benchmark. Additionally, the company is launching an internal drug discovery initiative called Claude Science, signaling a shift toward vertical industry integration.
DeepSeek has abandoned its non-commercial stance, securing $7 billion in funding to fuel massive hiring and expansion. The company is launching 'Harness AI', a coding agent designed to compete directly with Anthropic's Claude Code.
Cursor has launched 'Composer 2.5', a new AI model that achieves performance comparable to top-tier models like Claude Opus 4.7 and GPT-5.5 at 1/10th of the cost. It is currently ranked 3rd in Artificial Analysis's coding agent benchmarks.
Security researchers have identified a vulnerability called 'Agentjacking' where malicious actors can hijack AI coding agents using fake bug reports. This attack vector turns the agent against the user without requiring malware or stolen credentials.
DeepSeek is establishing a new 'Harness' team focused on AI coding agents. The initiative aims to directly compete with Anthropic's Claude Code, as confirmed by senior researcher Chen Deli.
Pi is the minimalist core framework powering OpenClaw, topping Terminal Bench 2.0 with under 1000-token prompts and just four tools: read, write, edit, bash. Creator Mario Zechner emphasizes letting users decide needs, avoiding bloated features like MCP or plan modes. It achieves high reliability by serializing state to files, bypassing context decay.
User releases Heretic/Uncensored INT4 quantized version of Qwen3.5 122B, faster than dense model on 256GB RAM cluster of ASUS Ascents and DGX Spark. Shares custom setups like 'The Cardinal' for agents and 'The Heretic' for speed. Emphasizes learning over hardware for coding tasks.
OpenAI executed 7 major Codex updates in 6 weeks, including desktop apps, GPT-5.3/5.4 models, Spark inference, and Security agent, surging usage 5x amid rivalry with Anthropic's Claude Code. Developers increasingly stack both for hybrid workflows.
Anthropic launches Remote Control for Claude Code, enabling smartphone or PC remote operation of local AI coding agents. Available as research preview for Pro and Max plan users.
Cursor has released an iOS application that allows developers to manage and interact with AI coding agents while away from their primary workstation. The app syncs with the desktop environment to enable session management and output review.