AI news and trends worldwide, updated nearly every day
Features, how-tos, and the latest developments on AI coding tools—Claude Code, Cursor, Cline, Windsurf, RooCode and more. Start from whichever tool you're curious about.
86 articles
On September 30, 2026, GitHub widened the HydraFusion research preview, which combines several models, to VS Code and the Copilot app. Billing is at each used model's standard rate, and the auto model selection discount does not apply. Here is how it works and when to reach for it.
OpenAI released GPT-6.1 Sol at DevDay 2026 on September 29, 2026. Input at $2 and output at $10 match GPT-6 Sol from a week earlier; the only change is cached input, from $0.20 to $0.10. Losing the none reasoning effort is the migration trap.
Anthropic released Claude Sonnet 5.5 on September 28, 2026. Per-million-token prices stay at Sonnet 5's $2 input and $10 output, while cost per task is down as much as 30% by the company's own testing. It is the first Sonnet model shipped with cyber safeguards.
Claude Code, Cursor and Google Antigravity compared on how you run agents in parallel and where you check the results. Based on each vendor's official documentation as consulted on September 25 and 26, 2026, covering parallel execution, cloud execution, the overview screen, review paths, scheduling, APIs/SDKs, and triggering from outside.
OpenAI released GPT-6 Sol and Luna on September 22, 2026. Sol is $2 input and $10 output; Luna is $0.10 and $0.50 — half of GPT-5.6's promotional pricing. Prompt caching also survives changes to reasoning effort and tool availability, which matters for the cost of running agents.
Anthropic released Claude Opus 5.5 on September 22, 2026. Input is $4 and output $20, 20% below Opus 5, with cache reads down 60%. At the same time, this is the first Opus model to ship with Fable 5.1-class safeguards, and most cybersecurity work is rerouted to Opus 4.8.
Anthropic put a rebuilt projects feature into beta in Claude Code on September 17, 2026. Set a goal and a repo and a coordinator carves up the work, with each thread opening PRs on a branch of its own. Because they run in parallel, Anthropic says usage limits arrive sooner.
The change to Claude Code's weekly usage limits that Anthropic announced in late August takes effect on September 14, 2026. It is a permanent 25% raise against the original baseline, but the temporary 50% boost ends the same day, leaving about 17% less than the level just before.
On September 10, 2026, Cursor released Projects in beta and began rolling it out to all users. Built for large units of work such as a feature or a migration, a coordinator agent that writes no code itself assigns work to subagents and keeps context in the cloud. No pricing is listed.
On September 11, 2026, the startup Accomplish published how opening an untrusted repository in Claude Code could let commands run outside the macOS sandbox without a permission prompt. It says it reported the issue on July 13 and that it was fixed in 2.1.247.
On September 6, 2026, OpenAI published data on coding agent use inside its research organization. As of mid-August, the median researcher used more than $600 a day at API prices, and the org as a whole used 3.1 agent-workdays per human workday. The post also quantifies how restrictions imposed after July 20 shifted GPU allocation.
OpenAI shipped Codex 0.153.0 on September 3, 2026 and 0.153.4 on September 4. The plugin CLI can now install from remote marketplaces, an experimental context-management setting arrives, and 0.153.4 makes Astra the bundled default model.
Anthropic released Claude Code 2.1.261 on September 4, 2026. It adds /skill-doctor, which shows which loaded skills go unused and what they cost in context, and changes auto mode to treat a link that packs content into a public diagram renderer's URL as an upload to that site.
GitHub made OpenAI's GPT-6 Astra generally available in Copilot on September 4, 2026. It covers Pro+, Max, Business and Enterprise, and can be picked from the model picker in ten places including VS Code and JetBrains IDEs. Billing is usage-based at provider list pricing.
The Contributor tier we covered in August now also lists the latest muse-spark-1.3, while its per-token prices and rate limits are unchanged. A second look at the gap against standard pricing, the Standard tier's explicit "not used to train" line, and the drop in rate limits.
A September 3 GitHub changelog entry announced that Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code and Claude Opus 4.7 will be deprecated across all GitHub Copilot experiences on October 2, 2026. Enabling the alternatives may be an administrator task.
In a September 2 changelog entry, Cursor described a setup that keeps cloud agents' tool execution entirely within your own network. It comes in two forms — My Machines for individuals and Team pools for organizations — and supports computer use on Linux and Mac.
GitHub announced on September 2 that enterprise-managed settings can set the default GitHub Copilot model for new conversations. Per-team overrides are configured via team-mappings.json, generally available for Copilot Business and Enterprise.
AI security company HiddenLayer announced a $100 million Series B on September 2, led by Delta-v Capital. Annual recurring revenue grew more than 10x over the past year, and the funding targets runtime protection for AI coding agents.
Google released Gemini 3.8 Flash and 3.8 Flash Cyber on September 2. Pricing stays at the 3.7 Flash introductory rate of $0.75/$3.75 per million tokens, but it expires December 31, 2026 and doubles. The cyber variant ships only to vetted Fairwind Program organizations.
OpenAI Codex 0.152.0 shipped on September 1. The planning tool is now off by default and needs a setting to turn on, and cloud task requests reject untrusted backend URLs to protect saved credentials.
Claude Code 2.1.257 shipped on September 1. It adds claude-fable-5-1 as the default Fable model and gives auto mode a rule that stops cloud metadata-credential fetches and cross-tenant reach from being auto-approved unless your environment marks them expected.
GitHub announced on September 1 that Copilot code review can approve pull requests. Approvals are disabled by default and controlled at the enterprise, organization, and repository level. Public preview.
Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1. Input and output stay at $10 and $50 per million tokens, but cache reads drop 75% to $0.25 per million, which the company estimates cuts typical workloads by about 25% and highly agentic ones by up to about 45%.
OpenClaw 2026.8.1, released on August 31, 2026, lets an agent request a credential without its value entering chat or model context, and adds automation permissions granted once per exact operation.
GitHub published a combined changelog on August 31, 2026 for Copilot in VS Code v1.132-v1.135, adding a side chat that shares the primary chat's context and prompt cache, and the ability to continue sessions created in other apps.
Debian's General Resolution on LLM usage has concluded, with "Responsible Use of Generative AI" winning out of eight options. It neither endorses nor prohibits the tools, and leaves review and legal accountability with the contributor.
GitHub announced on August 27, 2026 that Copilot code review can now review pull requests authored by bots and very large ones. The 300-file / 20,000-line limit no longer applies, and resolving a review comment now takes one of three reasons.
OpenAI released Codex CLI 0.151.0 on August 29, 2026. It preserves restored permission profiles across TUI turns, prevents /cd from weakening sandbox restrictions, and adds a configurable grace period for optional MCP servers.
The Claude Code CHANGELOG gained a 2.1.251 entry on August 28, 2026. It lists fixes around permissions, including file tools that could read or write outside the approved location via a swapped symlink, plus new hook events for model switching.
On August 28, 2026, OpenAI said it had notified SpaceX that it intends to wind down the contract providing OpenAI models to Cursor, proposing a shutoff date of November 12. It cited an inability to be confident SpaceX will stay within its terms of service, and said future models including Astra will not be provided.
Cursor published a cloud agent update on August 27, 2026. Agents can now start without a connected external SCM via "Start from scratch" and be saved to an Origin repo afterward, with a live browser preview of the agent's environment.
On August 28, 2026, GitHub previewed three separate changes to Copilot policies and billing. Business and Enterprise seats will require upfront payment, and chat data retention moves from 28 days to the life of the account as Copilot Chat and the cloud agent converge.
A stealth model with an anonymous provider, Ox Alpha, was added to OpenRouter free of charge on August 20, 2026 (UTC). Context length is 1,048,576 tokens with text, image and video input. The specs and price are verifiable, but the developer is undisclosed and the model page and the program EULA describe input handling differently.
Liquid AI published DSpark draft model checkpoints for three LFM2.5 models on August 20, 2026. The roughly 300M-parameter drafters deliver up to 3.18x throughput on an H100 and up to 2.87x on-device in the company's own measurements, with output identical under greedy decoding.
Salesforce and Slack launched Slack Code on August 20, 2026. Tagging a coding agent spins up a dedicated code channel where the whole team can watch the work, review diffs, run live previews, and approve before shipping. Claude, Devin, Copilot, ChatGPT, and Vercel agents are supported.
Cursor published a cloud agent update on August 19, 2026, adding Subscriptions that let agents wake on PRs, Slack threads, and scheduled tasks, a /goal command that holds an objective until it is met, and subagents running on their own virtual machines.
Cursor launched Origin, its own code hosting, in early beta on August 17, 2026. You can host new repos in Cursor or sync existing GitHub repos — but for synced repos, pushes still go to GitHub, which stays the source of truth.
Google released Gemini 3.7 Flash on August 13. Its workhorse model for coding and agents costs $0.75 per million input tokens and $3.75 per million output tokens through year-end, doubling on January 1, 2027. Here are the pricing structure and the benchmarks Google presented.
Cursor announced on August 13, 2026 that the Firetiger team is joining. Founded in 2024, Firetiger builds agents that watch software after it reaches production and feed findings back to coding agents. The standalone product is being sunset and signups are closed.
Cursor announced on August 14, 2026 that SpaceX's acquisition has closed, completing a process that began with an April model-training partnership with SpaceXAI. TechCrunch reports the deal at $60 billion in stock. What Cursor's own blog emphasizes is compute, not cash.
Cursor introduced builds for Cloud Agents on August 13, 2026. Agents boot into a ready environment prepared in the background instead of setting up each session. Internally, Cursor says environments boot 10x faster with 3x faster time to first token.
GitHub made xAI's Grok 4.6 available in GitHub Copilot on August 14, 2026. It covers Pro, Pro+, Max, Business, and Enterprise, selectable from the model picker across eight surfaces including VS Code and Copilot CLI. Business and Enterprise admins must enable it.
Z.ai released GLM-5.3 on August 14, 2026, keeping the same base model as GLM-5.2 and scaling only post-training. Exploitation benchmarks more than doubled over GLM-5.2, and the weights are held until safety evaluation completes, two weeks after launch.
DeepSeek has made DeepSeek-V4-Pro generally available and will change API pricing at 16:00 UTC on August 16, 2026. The new schedule has peak and off-peak rates, with off-peak at half of peak. v4-pro output goes from $0.87 to $3.96 at peak.
Anthropic's Frontier Red Team ran multiple Claude agents in shared environments. Coordinated swarms found 266 vulnerabilities vs 21 for independent agents, but conformity, collusion, and sabotage also emerged.
GitHub made Gemini 3.7 Flash available in Copilot on August 13, 2026 for Pro, Pro+, Max, Business, and Enterprise, billed at provider list pricing under usage-based billing. Org admins must enable the policy first.
GitHub made Agent Plugins 1.0 generally available on August 12 across VS Code, Copilot CLI, the Copilot SDK, and the Copilot app. The spec, published August 6 and initiated by Vercel, bundles Agent Skills with MCP servers — but the spec also states that vendor-namespace directories carry no portable semantics.
GitHub announced an August 11, 2026 update to GitHub Copilot for JetBrains: Copilot memory can retain and recall information across agent chat sessions, and Ollama is now usable as a BYOK provider. Server-based admin controls were expanded as well.
GitHub announced on August 11, 2026 that MAI-Code-1.1-Flash is available in Copilot, at a 73% lower list price than MAI-Code-1-Flash and a 0.25x premium request multiplier for annual subscribers. Business and Enterprise admins must enable it — the policy is off by default. The previous model retires September 10.
On August 7, 2026, GitHub reversed the behavior where enabling Code Quality created a ruleset that automatically requested a Copilot code review. The behavior had arrived with the July 20 general availability.
Anthropic released self-hosted environments for Claude Code in public beta on August 6, 2026, letting sessions run inside an organization's own network. Available to Claude Team and Enterprise organizations, off by default.
Anthropic will make auto mode the default for Claude Code on Pro, Max, and Team plans starting August 14. The company published a controlled study of 1,053 testers comparing human approval with its classifier. Enterprise stays opt-in for now.
Claude Code v2.1.224, published to npm on August 7, 2026, lets sessions send each other messages. Only plain text crosses over — never conversation history or files — and the feature runs on macOS and Linux.
GitHub said on August 7, 2026 that the Copilot impact dashboard now shows a Potential return on investment section with cost per developer per month, cost as a share of payroll, and pull requests per developer per month. Cost figures are estimates based on AI credit consumption.
GitHub said on August 7, 2026 that effort levels for Copilot code review are now generally available. Reviewers pick between Lite and Balanced per pull request, and the level that ran is now shown alongside the result.
GitHub announced on August 6, 2026 that the open-weight Kimi K3 model is generally available in Copilot. Business and Enterprise plans have it off by default, requiring admin enablement. Rollout is temporarily paused over a GitHub Actions incident; pricing will be $3 input, $15 output per 1M tokens.
Meta released Muse Code in beta on August 5, 2026, a terminal coding agent running on the new Muse Spark 1.2 model. Persistent async background agents, a crash-safe event log, and a Contributor tier at $0.10/$0.20 per million tokens in exchange for training rights.
GitHub announced Spark's upcoming deprecation on August 4, 2026. From that date it accepts no new users or apps, existing users have until August 31 to export, and the llm() function already stopped working.
Cursor's August 3, 2026 changelog adds three plugins covering Google Drive, Gmail, and Google Calendar — drafting and sending mail, creating and updating events, and creating and organizing files from the editor.
Alibaba's Qwen team officially released the 2.4-trillion-parameter Qwen3.8-Max on August 3, 2026, saying it will open-source Max-class weights for the first time. Pricing and benchmarks are now public — here is what the July preview left blank and what is still missing.
Cursor's July 29 changelog brings its iOS app to iPad on all paid plans: pinned sidebar chats for watching several agents, split-screen review, full file diffs, Apple Pencil markup, plus Bitbucket and Azure DevOps support.
Cursor launched Cursor Start, an India-exclusive plan at Rs649 (~$7) per month vs. $20 for Pro. Built on Cursor's own models like Grok 4.5 and Composer, with UPI payment support. Why India — 3M+ developers, its third-largest market — and what it signals ahead of the $60B SpaceX acquisition.
Cognition, the company behind coding agent Devin, acquired The Interaction Company, maker of the texting-native AI agent Poke, in a deal reported at low nine figures. Poke exchanged over 100 million messages in three months.
OpenAI's first branded hardware is Codex Micro, a $230 limited-run keypad built with Work Louder. Agent Keys glow with live agent status and a dial adjusts reasoning depth. What it does, and what it says about OpenAI's hardware ambitions amid the Apple lawsuit.
Anthropic published 'The Making of Claude Code' on July 6, 2026 - an oral history of how its coding agent was born. From a VS Code extension and an internal CLI tool called clide to a Slack demo that got 2-3 likes, here is the story behind the tool now writing 100% of its creator's code.
A side-by-side 2026 pricing comparison of six major AI coding tools - free tiers, individual, team/enterprise plans, and pay-as-you-go - plus how to pick the right one.
Meta Superintelligence Labs announced Muse Spark 1.1, an agentic coding model, on July 9, 2026, and opened the Meta Model API in public preview. It offers a 1-million-token context window through an OpenAI-compatible API, reportedly priced at $1.25 input / $4.25 output per million tokens, chasing Anthropic and OpenAI.
OpenAI released the GPT-5.6 family on July 9, 2026: flagship Sol ($5/$30), balanced Terra ($2.50/$15), and low-cost Luna ($1/$6) per million tokens. The launch introduces 'ultra', which coordinates four parallel agents, and an unusual hardware-passkey requirement tied to the models' cybersecurity capabilities.
SpaceXAI (formerly xAI) announced its flagship model Grok 4.5 on July 8, 2026. The first model trained jointly with Cursor, it's priced at $2/$6 per million tokens and claims 80 TPS speed with twice the token efficiency of leading models. Available now in Grok Build, Cursor, and the API; EU rollout expected mid-July.
Alibaba has reportedly classified Anthropic's Claude Code as 'high-risk software' and will ban internal use from July 10. The trigger: hidden code inside Claude Code that detected Chinese users. Combined with Anthropic's distillation accusations, the US-China AI rivalry is now reaching developer tools.
OpenAI announced on June 11, 2026 that it has agreed to acquire cloud development environment company Ona, integrating secure infrastructure that lets agents keep working in a customer's cloud even after the laptop is closed.
Cognition relaunched the AI coding editor Windsurf as Devin Desktop on June 2, 2026. The new Agent Command Center manages multiple agents on a Kanban board, while the legacy Cascade agent reaches end of life on July 1.
AI code editor Cursor acquires code review platform Graphite, aiming to deliver an integrated development experience from code creation to review and merge
Google's AI development environment 'Antigravity' announced in November 2025 adopts an agent-first design where AI agents autonomously write code. Powered by Gemini 3 and free to use, here's how it compares to Cursor and Claude Code
Cognition introduces SWE-1.5, a software engineering-focused AI model achieving up to 950 tokens/second—13x faster than Claude Sonnet 4.5.
Cline, the AI coding assistant for VS Code, releases major updates with Claude 4 family support, streaming features, SAP AI Core integration, and numerous enhancements
Engineers share Claude Code's internal structure (.claude directory, JSONL logs) and practical development usage
Starting as a VS Code fork, AI code editor Cursor reached nearly $10 billion valuation in just 2 years. With 2x productivity gains, it's winning massive support from developers worldwide
Rebranded from Codeium, Windsurf AI Editor captivated 1 million developers in just 4 months. Discover the future of development experience envisioned by MIT founders
Cline is an open-source AI coding assistant for VS Code that can autonomously write code, edit files, and run terminal commands. How founder Saoud Rizwan's tool passed 4.8 million installs and what sets it apart.
RooCode was a free, open-source VS Code extension where multiple AI agents collaborated like a dev team. The extension was shut down on May 15, 2026 and its GitHub repository archived; this guide is kept as a record of how it worked, with migration options noted.
Practical patterns for working with Claude Code, drawn from Anthropic's official best practices and quickstart documentation
As Cursor breaks through $500M annual revenue, insights from discussions between Anthropic and Cursor executives reveal the current state and future of AI coding. Examining Claude 4's innovative capabilities and the changing role of developers.
Anthropic announces Claude 4 Opus and Sonnet 4. Achieving over 72% on SWE-bench with parallel tool execution and extended thinking capabilities. Enterprise memory features also added.
Claude 4, released by Anthropic in May 2025, has garnered attention as the world's best coding AI with a 72.5% SWE-bench score. A detailed explanation of its innovative features and practicality including hybrid reasoning, 7 hours of autonomous coding, and full Japanese support
Last updated: 2026-10-02
The timeline data on this page is available under the Creative Commons Attribution 4.0 International (CC BY 4.0) license, for commercial and non-commercial use alike. Please credit the source with the site name, the URL of this page, and the confirmation date of the data.
Example: Wain, "AI Coding Tools Timeline," confirmed 2026-10-02, https://ai.wain.blog/en/coding-tools/