ai-powered-markdown-translatorTranslated article from French to English with gpt-5.4-mini.
May 18, 2026 marks Anthropicโs acquisition of Stainless โ the SDK generator behind all official Claude API libraries. GitHub Copilot simultaneously moves to GPT-5.3-Codex, the first LTS model guaranteed for 12 months, while remote control of CLI sessions reaches general availability on mobile, web, and IDEs. At the same time, Qwen3.7 Preview arrives on Arena, xAI opens Hermes to X Premium subscribers, and FLUX launches its official MCP server.
Anthropic acquires Stainless, SDK and MCP server generator
May 18, 2026 โ Anthropic announces the acquisition of Stainless, a company founded in 2022 by Alex Rattray and specialized in the automatic generation of SDKs (Software Development Kits) and MCP server tooling (Model Context Protocol).
Stainless is already the quiet engine behind all official Anthropic API SDKs. In practical terms, the company turns an API specification into ready-to-use libraries for TypeScript, Python, Go, Java, Kotlin, and other languages โ each designed to integrate naturally into its ecosystem. Hundreds of companies also rely on Stainless to generate their own SDKs, command-line tools, and MCP connectors.
| Aspect | Detail |
|---|---|
| Founded | 2022, by Alex Rattray |
| Languages covered | TypeScript, Python, Go, Java, Kotlin, and more |
| Current clients | Hundreds of companies (SDKs, CLI, MCP servers) |
| Anthropic integration | All official Claude API SDKs already generated by Stainless |
โAgents are only as useful as what they can connect to. Weโre excited to bring the Stainless team into Anthropic to advance Claudeโs ability to connect to data and tools.โ โ [Katelyn Lesse, Head of Platform Engineering, Anthropic](url unchanged)
Founder Alex Rattray highlights the natural fit of the deal:
โI started Stainless because SDKs deserve as much care as the APIs they wrap.โ โ [Alex Rattray, Founder and CEO of Stainless](url unchanged)
This acquisition fits directly into Anthropicโs MCP strategy: the protocol was designed to make agent connectivity possible, and Stainless is the tool that generates connectors at scale. By absorbing the team, Anthropic strengthens its control over the entire agent tooling chain, from models all the way to SDKs.
GPT-5.3-Codex becomes the default LTS model for Copilot Business and Enterprise
May 17, 2026 โ GitHub announces that GPT-5.3-Codex is now the new default model for all Copilot Business and Copilot Enterprise organizations, replacing GPT-4.1. It is the first LTS model (Long-Term Support) for GitHub Copilot, developed in partnership with OpenAI.
The LTS guarantee is concrete: availability is ensured for a full 12 months from launch, until February 4, 2027 (the model launched on February 5, 2026). Organizations therefore do not have to endure unplanned model changes in the middle of a development cycle.
| Aspect | GPT-5.3-Codex (new) | GPT-4.1 (old) |
|---|---|---|
| Status | LTS default model | Forcefully active until June 1, 2026, then deprecated |
| Premium request multiplier | 1x | 0x (free until June 1) |
| Availability guarantee | 12 months (until Feb. 4, 2027) | No LTS guarantee |
| Code survival rate | Significantly high (GitHub Copilot data) | โ |
| Affected plans | Business, Enterprise | Same |
| Unaffected plans | Pro, Pro+, Free | Same |
Note for administrators: GPT-4.1 remains active at 0x multiplier until June 1, 2026, when usage-based billing begins. Individual plans (Pro, Pro+, Free) are not affected by this timeline โ they follow the standard deprecation schedule.
๐ GitHub Changelog
Remote control Copilot CLI reaches general availability โ mobile, web, VS Code, and JetBrains
May 18, 2026 โ Remote control of Copilot CLI sessions reaches general availability on GitHub Mobile and github.com. This major update greatly expands the covered surfaces and removes previous restrictions on repository types.
Whatโs new in this GA release:
| Feature | Detail |
|---|---|
| Non-GitHub repository support | Sessions on any repository, including those not hosted on GitHub |
| Repository-less directories | Accessible at github.com/copilot/agents |
| VS Code | Enable github.copilot.chat.cli.remote.enabled, start from the Chat view |
| JetBrains | Remote control available from the IDE |
| CLI startup | copilot --remote or /remote on during a session |
| Mobile connection | Scan the QR code to open on GitHub Mobile |
In practice, a developer can now launch a Copilot agent in a local terminal, scan a QR code, and monitor or control its progress from their phone โ whether for a GitHub repository, a private project, or any directory.
Requirements for Business/Enterprise organizations: administrators must enable the remote control and CLI policies in the organizationโs Copilot settings.
๐ GitHub Changelog
Anthropic โ Claude developer tooling
Claude Design โ token limits doubled
May 18, 2026 โ Anthropic announces a doubling of token limits for Claude Design, the visual creation product launched by Anthropic Labs in April 2026. This increase applies to all subscriptions: Free, Pro, Max, Team, and Enterprise.
Claude Design makes it possible to create designs, prototypes, slides, and visual documents in collaboration with Claude. According to the official announcement from the @claudeai account, limits have been doubled across every subscription โ offering more room for complex projects or long iterations.
Fast mode Claude Code โ switch to Opus 4.7
May 18, 2026 โ @ClaudeDevs announces that Fast mode in Claude Code now switches by default to Claude Opus 4.7. This update brings Anthropicโs latest model into the accelerated workflow.
| Aspect | Value |
|---|---|
| Fast mode model | Claude Opus 4.7 |
| Speed gain | ~2.5ร compared to standard mode |
| Quality | Same as standard Opus |
| Activation | /fast in Claude Code |
| Pricing | Higher per token than standard mode |
Prompt cache diagnostics in Claude Console
May 18, 2026 โ Anthropic adds a prompt cache diagnostics feature to Claude Console. When a request misses the cache, the console now precisely identifies which part of the prompt changed and how many additional tokens that cost.
This tool is aimed at developers optimizing their API costs through prompt caching. A cache miss triggers full billing for context tokens โ the diagnostics help identify and correct unintended variations. The feature is available in the /usage/cache section of Claude Console, with documentation at platform.claude.com/docs/en/build-with-claude/cache-diagnostics.
GitHub Copilot โ New APIs
Copilot Spaces API reaches general availability
May 18, 2026 โ The GitHub Copilot Spaces API exits preview and reaches general availability. Teams can now create, read, update, and delete Spaces programmatically from their own applications, and manage associated collaborators and resources.
This evolution is especially useful for companies managing many Spaces and wanting to automate these operations without going through the GitHub web interface.
๐ GitHub Changelog
Audit Copilot cloud agent configuration via REST API
May 18, 2026 โ A new public preview REST API makes it possible to audit the Copilot cloud agent configuration for each repository programmatically. It returns configuration details: enabled MCP servers, allowed tools, GitHub Actions workflow policy, and firewall configuration.
Important clarification: this API is in public preview (not yet generally available). It is ideal for security teams that want to inspect the posture of their repositories at scale.
๐ GitHub Changelog
Agents โ Manus Scheduled Tasks 2.0
May 18, 2026 โ Manus (now integrated into Meta) launches Scheduled Tasks 2.0, a redesign of recurring task scheduling in the agent. The key new feature: scheduled runs can now stay within the context of the original task โ same conversation, same files, same history โ instead of starting from a blank slate.
Three distinct improvement areas:
| New feature | Detail |
|---|---|
| Persistent context | Scheduled runs resume in the same task, with access to the same history and files |
| Integrated web apps | Apps created with Manus can include their own schedules (data refresh, reports, reminders) |
| Manus Projects integration | Reuse of shared configuration: files, skills, connectors, instructions |
| Calendar and scheduling views | Visual tracking of upcoming runs and history |
Available immediately to all users. Advanced controls include Run options (same task or new), Skip confirmations for trusted workflows, and agent selection.
๐ Manus Blog
Media โ NVIDIA, Runway, and FLUX
NVIDIA OpenShell v0.0.43
May 18, 2026 โ NVIDIA releases OpenShell v0.0.43, the new version of its framework for AI agents. This update focuses on three areas: authentication, bidirectional streaming, and sandbox isolation.
| Feature | Detail |
|---|---|
| Bidirectional TTY | Real-time streaming in both directions |
| OIDC auth TUI | OIDC authentication directly in the interactive terminal |
| Decoupled HTTPS and mTLS | Independent transport-layer configuration |
| TOML config (RFC 0003) | Gateway configuration format via TOML |
| ext4 sandboxes | Sandboxes start from ext4 disks |
| DNS removed from mapper | Removal of DNS in the sandbox mapper to block data exfiltration |
NVIDIA GTC Berlin 2026
May 18, 2026 โ NVIDIA announces the Berlin edition of its GTC conference, scheduled for October 20โ22, 2026. The event will feature in-depth technical sessions, hands-on workshops, and presentations on next-generation AI infrastructure, aimed at developers, researchers, and decision-makers. Registration is not yet open.
Runway Characters โ tool calling
May 18, 2026 โ Runway expands its Characters โ real-time video agents โ with the ability to call tools. Previously limited to speech, Characters can now trigger programmatic actions at the userโs request. This opens the door to interactive video agents capable of performing tasks in third-party applications.
Official FLUX MCP โ images in Claude, Cursor, and Hermes
May 18, 2026 โ Black Forest Labs launches the official MCP (Model Context Protocol) server for FLUX. This integration makes it possible to generate FLUX images directly from Claude, Cursor, Hermes, and other MCP-compatible agent environments. The server automatically selects the right FLUX model based on context. Available for free at mcp.bfl.ai.
๐ @bfl_ml announcement
Models and agents โ Qwen3.7 Preview and xAI Hermes
Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview on Arena
May 18, 2026 โ Alibaba announces the arrival of two preview models on the Arena platform: Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview. This preview marks a notable step forward for Alibaba in global rankings.
| Model | Overall ranking | Strengths |
|---|---|---|
| Qwen3.7-Max-Preview | #13 in Text Arena | #7 Math, #9 Expert, #9 Software & IT, #10 Coding |
| Qwen3.7-Plus-Preview | #16 in Vision Arena | โ |
| Alibaba (lab) | #6 in Text, #5 in Vision | Significant progress from outside the top 10 |
The Qwen3.7 series is not yet available for download โ the official qwenlm.github.io blog does not yet show a dedicated post. The team says an official release is imminent.
๐ @Alibaba_Qwen announcement
Hermes Agent opens to X Premium subscribers and X search
May 16, 2026 โ xAI expands Hermes Agentโs capabilities with two new features: X Premium subscriptions are now accepted (in addition to Grok subscriptions), and the agent can perform searches in X posts in real time. To enable the integration, users must connect their X account in grok.com.
This update generated 3.8 million views in a few hours, confirming massive interest in the open-source personal agent Hermes from Nous Research, which runs continuously on any server and retains persistent memory between sessions.
๐ @xai announcement
OpenAI + Dell Technologies โ Codex in hybrid and on-premises environments
May 18, 2026 โ OpenAI and Dell Technologies team up to let companies deploy Codex in their hybrid and on-premises environments, where their critical data and systems are already hosted.
Codex is now one of OpenAIโs fastest-growing enterprise products, with more than 4 million active developers every week. The product goes beyond software development alone: teams use it to prepare reports, route product feedback, qualify leads, and coordinate business workflows.
The partnership connects Codex to the Dell AI Data Platform so enterprises can bring the agent closer to their internal codebases, documentation, operational systems, and workflows. ChatGPT Enterprise and other API-based solutions will also be included.
โCollaborating with OpenAI brings together Dellโs industry-leading enterprise grade infrastructure with cutting edge agentic AI harnesses and models from OpenAI. The Dell AI Factory with OpenAI Codex will allow enterprises to deploy AI where enterprise data already lives, within their premises, giving customers a practical, secure path to deploying AI agents at scale.โ โ [The collaboration with OpenAI brings together Dellโs cutting-edge enterprise infrastructure and OpenAIโs advanced agentic capabilities. Dell AI Factory with OpenAI Codex will enable companies to deploy AI where their data already lives, on their own infrastructure, giving customers a practical and secure path to deploying AI agents at scale.](url unchanged)
๐ OpenAI announcement
Briefs
- Code with Claude London (May 19, 2026) โ Second edition of Anthropicโs developer tour in London on May 19, with technical sessions, demos, and discussions with the Claude teams. Livestream available. ๐ source
- Google I/O 2026 โ live stream on May 19 โ Google DeepMind announces the Google I/O keynote on May 19 at 10 a.m. PT (7 p.m. Paris time), live on X. ๐ source
What it means
The acquisition of Stainless consolidates Anthropicโs MCP chain end to end. By absorbing the SDK generator that was already its official provider, Anthropic is not changing its tools โ it now owns them. The direct result: SDKs that are updated faster with every Claude API change, and likely accelerated MCP support across all languages. It is as much a defensive move as an offensive one: competitors can no longer use Stainless to generate SDKs as tightly integrated with the Claude ecosystem.
Copilot LTS marks a maturity shift for enterprise AI. The 12-month guarantee without unplanned model changes addresses a real need for CIOs and engineering teams: being able to certify, test, and validate behavior without fearing that a silent update will break their workflows. The 1x multiplier โ compared with GPT-4.1โs 0x โ means a real cost for organizations, but in exchange for predictability that has value in production. This is the first LTS; others will likely follow.
The convergence of agents toward shared infrastructure is accelerating. Black Forest Labsโ official FLUX MCP, remote control of Copilot CLI sessions, and Manus Scheduled Tasks 2.0 all illustrate the same movement: agents are no longer isolated tools but components of an interconnected ecosystem. FLUX integrates MCP to plug into Claude, Cursor, and Hermes. Copilot CLI becomes controllable from any surface. Manus retains context between scheduled runs. The MCP standard and the remote-agent architecture are gradually emerging as the basic primitives of this ecosystem.
The OpenAI+Dell partnership and the arrival of Qwen3.7 point to two tensions in the enterprise market. On one side, there is growing demand for on-premises deployment for sensitive data โ Dell is meeting that need with certified infrastructure. On the other, the rapid rise of Asian labs: Alibaba at #6 worldwide in text and #5 in vision with an Arena preview illustrates that benchmark competition is now open competition, with the official release of Qwen3.7 imminent.
Sources
- Anthropic acquires Stainless โ Official announcement
- GPT-5.3-Codex new base model for Copilot Business/Enterprise โ GitHub Changelog
- Remote control of Copilot CLI now generally available โ GitHub Changelog
- Claude Design โ token limits doubled โ @claudeai on X
- Fast mode in Claude Code moves to Opus 4.7 โ @ClaudeDevs on X
- Prompt cache diagnostics in Claude Console โ @ClaudeDevs on X
- Copilot Spaces API now generally available โ GitHub Changelog
- Audit Copilot cloud agent configuration via REST API โ GitHub Changelog
- Manus Scheduled Tasks 2.0 โ Manus Blog
- NVIDIA OpenShell v0.0.43 โ @NVIDIAAI on X
- NVIDIA GTC Berlin 2026 โ @NVIDIAGTC on X
- Runway Characters Tool Calling โ @runwayml on X
- Official FLUX MCP โ @bfl_ml on X
- Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview on Arena โ @Alibaba_Qwen on X
- Hermes Agent โ X Premium and X search โ @xai on X
- OpenAI + Dell Technologies partnership for Codex โ OpenAI
- Code with Claude London โ @claudeai on X
- Google I/O 2026 teaser โ @GoogleDeepMind on X