SWE-Explore finds coding agents miss the key lines
A Shanghai Jiao Tong University-led benchmark shows agents often find the right files but identify the exact bug-relevant lines only about 14% to 19% of the time.
AI Coding News and Analysis · Page 2
162 verified stories covering AI Coding, product updates and industry developments. Page 2 of 2
Zhipu opens GLM-5.2 with a 1M-token context
Zhipu's GLM-5.2 pairs a 744B-parameter MoE design with a 1M-token window and a coming MIT-style open-weight release for coding agents.
Kimi K2.7 Code undercuts GPT-5.5 on coding token prices
Moonshot's open MoE model offers far cheaper input and output tokens while improving agent-style coding benchmarks over K2.6.
Cursor Auto-review cuts agent interruptions to 7%
Cursor's classifier lets low-risk agent actions proceed while pausing higher-stakes steps, reducing approval fatigue for developers.
Claude Code adds /fork for parallel AI coding experiments
Anthropic's new command copies a live coding session so developers can try risky approaches without losing the original context.
Moonshot open-sources Kimi K2.7-Code for AI coding
Moonshot open-sources Kimi K2.7-Code for AI coding.
OpenAI buys Ona to strengthen Codex cloud workspaces
The acquisition adds cloud development infrastructure to Codex, tightening OpenAI’s push from coding assistant toward full software production workflows.
OpenAI acquires Ona to deepen Codex infrastructure
OpenAI’s purchase of the Gitpod-linked cloud sandbox company points Codex toward managed development environments, not just code suggestions.
Cohere releases North Mini Code for local coding
Cohere’s 30B-parameter MoE model activates about 3B parameters, supports 256K context and is released under Apache 2.0.
Cursor chooses London for its European base
Cursor plans a London European headquarters and wants to grow its regional team to about 200 people by year-end.
Supabase raises $500 million as agents build databases
The Postgres-based backend platform says its user base has more than doubled since the last round and agents now deploy most databases on the platform.
GitHub Copilot metered billing sparks backlash
Copilot’s AI Credits make agentic sessions and frontier models usage-based, exposing developers to bills far above the old flat subscription.
OpenAI pushes Codex beyond programmers into every office role
New role plugins, Sites and annotations turn Codex from a coding assistant into a broader workflow tool for non-developers.
Google quietly offers to buy private app code for AI training
A reported Google Play pilot asks developers to sell source code, including dormant projects, while the email itself avoids saying AI.
Microsoft unveils seven in-house MAI models, led by a coding flash model
At Build 2026, Microsoft presented seven self-trained models, with MAI-Code-1-Flash beating Claude Haiku 4.5 on its cited coding benchmarks.
GitHub Copilot moves from flat fees to token-based billing
GitHub is replacing Copilot premium requests with AI Credits, making heavier agentic coding sessions pay closer to their actual token cost.
Microsoft Project Polaris will replace GPT inside Copilot in August
At Build 2026, Microsoft positioned Windows as an agent platform and said Project Polaris will become the default GitHub Copilot reasoning engine.
Sekai turns vibe coding into 15 million consumer mini apps
Sekai has raised a $20 million Series A after users generated 15 million mini apps and spent more than an hour a day on the mobile platform.
GitHub Copilot’s usage billing shocks heavy agent users
Copilot is moving from premium request counts to AI Credits, exposing the real token cost of agent-style coding.
Cognition's valuation jumps to $26B in eight months
AI coding startup Cognition raises over $1 billion at a $26 billion valuation, claiming 90% of its internal code is now generated by its own AI agent Devin.
Anthropic survey: Most developers merged Claude-written PRs without review
At Anthropic's first European developer event, a show of hands revealed that most attendees had merged pull requests written entirely by Claude without reading the code.
Cursor Integrates Automations Directly Into IDE
Cursor's 3.5 update moves its Automations feature from a standalone web page into the IDE's Agents Window, enabling multi-repo support and agent-driven workflows beyond coding.
Google Antigravity 2.0 lets 93 AI agents build an OS in 12 hours
Google launched Antigravity 2.0 at I/O 2026, a standalone agent platform that can orchestrate 93 sub-agents to build an operating system in 12 hours.
Cursor launches Composer 2.5, matching top models at a tenth of the cost
Cursor's third-generation coding model Composer 2.5 matches Opus 4.7 and GPT-5.5 on key benchmarks while costing only a tenth per task.
Gemini 3.5 Flash beats GPT-5.5 in agentic benchmarks, costs just $1.5
Google's Gemini 3.5 Flash outperforms GPT-5.5 on agentic benchmarks while being dramatically cheaper, and becomes the default model across Gemini app and Search.
Codex comes to the ChatGPT mobile app
OpenAI added Codex controls to ChatGPT on iPhone, iPad and Android, letting users monitor and approve coding tasks away from the desktop.
Cursor cloud agents learn multi-repo pull requests
Cursor’s cloud agents now support multi-repository environments, faster builds and enterprise controls for teams with microservice codebases.
Coder Brings Coding Agents On Prem
Coder Agents gives enterprises a self-hosted, model-agnostic way to run AI coding agents without sending code or prompts outside their network.
Cognition in Talks for $25 Billion Valuation
Cognition AI, the maker of the AI software engineer Devin, is reportedly in funding talks at a $25 billion valuation, a sixfold increase in 13 months.
Google CEO says 75% of new code is now written by AI
Sundar Pichai revealed during a quarterly earnings call that 75% of Google's new code is now AI-generated and human-reviewed, up from 25% 18 months ago. The shift is reshaping developer roles and highlights internal use of rival tools like Claude Code at DeepMind.
DeepSeek V4 Launches: 1.6 Trillion Parameters, 8x Cheaper Than GPT-5.5
DeepSeek released V4-Pro and V4-Flash under MIT license, achieving top competitive programming scores while pricing output at $3.48/M — a fraction of GPT-5.5 and Claude Opus 4.7.
Developer Survey: Claude Code Usage Grows Sixfold in Six Months
A new JetBrains survey of over 10,000 professional developers reveals that 90% now use AI tools, with Claude Code surging from 3% to 18% adoption in six months while GitHub Copilot's growth stalls.
Claude Code Desktop Adds Parallel Sessions and Routines for Asynchronous Work
Anthropic redesigned the Claude Code desktop app with multi-session management, built-in dev tools, and Routines for automated, event-triggered coding tasks.
Tokenmaxxing Is Dragging Down Developer Productivity
AI coding tools have been widely adopted for two years, but a new phenomenon called "Tokenmaxxing" is emerging where developers treat AI token budgets as a status symbol. Data shows code churn is up to 9.4 times higher, and only 10-30% of AI-generated code survives after a few weeks.
Alibaba open-sources Qwen3.6-35B, achieving 73.4% on SWE-bench
Alibaba's Qwen team has released Qwen3.6-35B-A3B under Apache 2.0, a 35B-parameter MoE model that activates only 3B per token, scoring 73.4% on SWE-bench Verified and outperforming Google's Gemma 4-31B by 21 points.
Cursor in Talks for $50 Billion Valuation in New Funding Round
AI coding startup Cursor is negotiating a $2 billion funding round at a pre-money valuation exceeding $50 billion, nearly doubling its valuation in six months.
AI boosts PR merges by 98%, but delivery hasn't sped up
Agoda's engineering team found that while AI coding tools significantly increased individual developer output, project delivery speed remained unchanged, with code review becoming the new bottleneck.
AI coding tools layer: Cursor orchestrates, Claude writes code, Codex reviews
In the first week of April 2026, Cursor v3, OpenAI's Codex plugin for Claude Code, and an emergent three-layer stack show that AI coding tools are diverging into specialized layers rather than converging into one.
Apple Sends Siri Engineers to AI Coding Bootcamp to Learn Claude Code
Apple is sending Siri team engineers to an AI coding bootcamp focused on using tools like Claude Code, as the company scrambles to overhaul Siri ahead of WWDC in June.
Factory raises $150M at $1.5B valuation for enterprise AI coding
Factory has closed a $150 million Series B led by Khosla Ventures, reaching a $1.5 billion valuation, with Morgan Stanley, EY, and Palo Alto Networks already paying customers for its enterprise-grade AI coding agent platform.
GLM-5.1 Tops SWE-Bench Pro, Trained Entirely on Huawei Ascend
Z.ai (formerly Zhipu AI) released GLM-5.1, a 754B-parameter open-source MoE model that scored 58.4% on SWE-Bench Pro, surpassing GPT-5.4 and Claude Opus 4.6. The model was trained entirely on Huawei Ascend 910B chips with MindSpore, using zero NVIDIA hardware.
OpenAI runs Codex-Spark on Cerebras, 15x faster
OpenAI has deployed its GPT-5.3-Codex-Spark model on Cerebras WSE-3 chips instead of Nvidia GPUs, achieving over 1,000 tokens per second — roughly 15 times faster than the standard version. The move marks OpenAI's first production deployment on non-Nvidia hardware.
OpenAI turns Codex into a desktop automation agent
OpenAI has updated Codex with background computer use, allowing it to operate other macOS apps autonomously while users continue working. The update also includes a built-in browser, image generation, over 90 new plugins, SSH access, and self-scheduling capabilities.
TypeScript Tops GitHub as the #1 Programming Language
GitHub's Octoverse 2025 data shows TypeScript became the most contributed-to language in August 2025, driven in part by AI tools creating a self-reinforcing "convenience loop."
Claude Opus 4.7 triples coding performance in production bug fixes
Anthropic released Claude Opus 4.7 on April 16, delivering a threefold improvement in production code repair tasks over its predecessor and scoring 64.3% on SWE-bench Pro, surpassing GPT-5.4 and Gemini 3.1 Pro.
Kimi K2.6 Code Launches, Priced 5x Cheaper Than Claude Sonnet
Moonshot AI has rolled out the K2.6 code preview to all Kimi Code subscribers, offering deeper reasoning chains, improved multi-step agent planning, and more reliable tool calls at a fraction of the cost of competing models.
VS Code shifts from monthly to weekly releases with AI agents
Microsoft's VS Code team accelerated its release cycle from monthly to weekly by leveraging AI agents, doubling code commits and issue closures without expanding the team.
Cursor 3.0 Redefines the IDE as an Agent Manager
Cursor 3.0 introduces a new Agents Window for parallel task execution, a Design Mode for visual UI targeting, and built-in /worktree sandbox support, shifting the developer's role from executor to orchestrator.
Sonnet 4.6 trails Opus by just 1.2 points in coding at 5x lower price
Anthropic's Claude Sonnet 4.6 matches Opus 4.6 on coding and computer use benchmarks while costing five times less, making it the smarter choice for most developers and enterprises.
Spotify’s top engineers haven’t written a line of code since December
Spotify co-CEO Gustav Söderström revealed that the company’s best engineers have not written any code since December, thanks to an internal AI system called Honk that integrates Claude into a structured development pipeline. The approach has accelerated deployment and enabled over 50 new features in 2025, shifting engineers’ focus from writing code to defining requirements and overseeing quality.
GitHub Copilot CLI Reaches General Availability, SDK Now in Public Preview
GitHub has released Copilot CLI as generally available for all paid Copilot users, adding MCP server management, a Critic agent, dual Plan and Autopilot modes, and multiple model choices. Separately, the Copilot SDK is now in public preview, exposing the same agent runtime for embedding into custom applications.
Qodo raises $70M for code verification as AI coding scales
Israeli startup Qodo has closed a $70 million Series B, bringing its total funding to $120 million, to provide an independent verification layer for AI-generated code.
Cursor Flagship Product Uses Chinese AI Model
Cursor's new Composer 2 model is built on Moonshot AI's open-source Kimi K2.5, a fact the company initially omitted from its launch materials. The revelation has sparked debate about transparency in AI development.
Alibaba's Qwen 3.6 Plus Supports 1M Context, Rivals Claude Opus in Agentic Coding
Alibaba Cloud's Qwen 3.6 Plus offers a 1M-token context window for repository-level code analysis, matches Claude Opus 4.5 on SWE-bench, and generates front-end code from UI screenshots.
Vibe coding makes software feel conversational, but it is not a shortcut around judgment
Vibe coding turns natural-language prompts into working code. It is useful for prototypes and routine tasks, but still demands engineering judgment when reliability, security or long-term maintenance matter.
METR Study: AI Tools Slowed Experienced Programmers by 19%
A study by AI safety research organization METR found that experienced developers using AI coding tools completed tasks 19% slower on average, challenging assumptions about productivity gains.
GPT-5.3 Codex Moves From Code Completion to Real-Time Collaboration
OpenAI's Codex has evolved from a code completion tool into a real-time collaborative coding agent, with new Spark and Realtime modes, while also raising critical security concerns that prompted industry-wide reassessment.
Copilot at $10 a Month: What Value Does It Really Offer?
GitHub Copilot's $10/month pricing stands out in 2026 as AI coding tools generally rise in cost. This article examines its three-tier capabilities, compares it with rivals like Claude Code and Cursor, and outlines who benefits most from the plan.
OpenAI Shifts Codex Billing to Per-Seat, Usage-Based Model
OpenAI has introduced usage-based billing for dedicated Codex seats alongside a price cut for ChatGPT Business, signaling a push to mature AI coding tools for enterprise deployment.
How to Choose Among the Top Three AI Coding Tools in 2026
Claude Code, Cursor, and GitHub Copilot have taken three distinct paths. This article compares their positioning, capabilities, developer sentiment, and pricing to help you decide.
Claude Code terminal tool gains traction among developers
Claude Code is a command-line AI coding tool that works directly in the terminal with no GUI, leveraging a 100K-token context window and parallel agent teams for large-scale refactoring.
What SWE-bench Actually Measures
SWE-bench has become the de facto benchmark for AI coding ability, but understanding what it tests—and what it doesn't—is crucial for interpreting the scores.