Dateline: May 8, 2026 | Next update: May 15, 2026
The biggest story of the week is infrastructure: Anthropic struck a compute deal with SpaceX's Colossus 1 data center and immediately passed the benefit to users by doubling Claude Code rate limits and removing peak-hour caps. That announcement headlined the "Code with Claude 2026" developer event in San Francisco on May 6. Also this week: Managed Agents gained three major capabilities (dreaming, outcomes, and multiagent orchestration); connectors expanded to over 200 apps including Uber, Spotify, and Instacart; and a new enterprise AI services company was announced in partnership with Blackstone, Hellman & Friedman, and Goldman Sachs.
Claude / Anthropic
SpaceX compute deal — Claude Code limits doubled
Anthropic signed an agreement with SpaceX to use all of the compute capacity at SpaceX's Colossus 1 data center in Memphis, Tennessee. This adds more than 300 megawatts of new capacity — over 220,000 NVIDIA GPUs — coming online within the month. Anthropic immediately passed the benefit to paid subscribers with three concrete changes to Claude Code and API limits.
(1) Claude Code five-hour rate limits doubled for Pro, Max, Team, and seat-based Enterprise plans. (2) Peak-hour rate limit reductions removed entirely for Pro and Max — no more slower limits during busy periods. (3) Claude Opus API rate limits materially raised for heavy API users. Free plan is not affected. Anthropic also expressed interest in partnering with SpaceX on orbital AI compute capacity as a longer-term ambition.
Colossus 1 data center: Memphis, TN | 300+ MW, 220,000+ NVIDIA GPUs | Joins existing deals: Amazon (up to 5GW, 1GW by end 2026), Google/Broadcom (5GW from 2027), Microsoft/NVIDIA ($30B Azure), Fluidstack ($50B US infrastructure) | Free plan: unaffected
Best for: All paid Claude Code subscribers — especially Pro users who hit limits frequently
'Code with Claude 2026' — developer event highlights
Anthropic held its first major developer event of the year on May 6, anchored by CPO Ami Vora and Claude Code creator Boris Cherny. Key stats shared: API volume is up 17x year-on-year. Shopify and Mercado Libre (23,000 engineers) were highlighted as flagship Claude Code customers; Mercado Libre is targeting 90% autonomous coding by Q3 2026. No new model was announced — the event was framed explicitly as being about making existing products work better.
API volume up 17x year-on-year disclosed publicly for the first time. CI auto-fix confirmed — Claude Code automatically files fixes against PRs in CI. Code Review now used by every team at Anthropic internally. Remote Agents (control your laptop from your phone) demonstrated live on stage. No new model announced.
CI auto-fix: files automatic PR fixes | Code Review: parallel multi-agent cloud review (launched Week 16) | Remote Agents: phone-to-laptop remote control via Cowork | Event recap available on Anthropic's YouTube channel
Best for: Developers, enterprise engineering leaders, Claude Code power users
Claude Managed Agents — dreaming, outcomes, and multiagent orchestration
Claude Managed Agents gained three major new capabilities this week, each addressing a different limitation of long-running agentic work. Netflix is already using multiagent orchestration in production for its platform team.
Dreaming (research preview): a scheduled process that reviews past agent sessions and memory stores, extracts patterns, and automatically curates memories so agents self-improve over time. You control how much autonomy dreaming has — it can update memory automatically, or surface changes for your review first. Outcomes: define a goal, let the agent run, and receive a webhook notification when it's done — with improved file generation quality (+8.4% on DOCX, +10.1% on PPTX in internal benchmarks). Multiagent orchestration: a lead agent breaks a job into pieces and delegates each to a specialist subagent with its own model, prompt, and tools. Specialists work in parallel on a shared filesystem; every agent remembers what it has done. Full trace visibility is available in the Claude Console.
Dreaming: scheduled memory review and self-improvement loop; configurable auto-update or human-review modes | Outcomes: webhook on completion; file gen improvement: +8.4% DOCX, +10.1% PPTX | Multiagent: lead agent + parallel specialist subagents, shared filesystem, persistent events, full Claude Console tracing | All under managed-agents-2026-04-01 header
Best for: Enterprise teams building self-improving agents, long-horizon document workflows, parallel investigative tasks
Claude connectors — now 200+ apps including Uber, Spotify, Instacart
The Claude connector directory crossed 200 apps this week, expanding well beyond workplace tools into everyday consumer apps. New connectors include AllTrails, Instacart, Audible, Tripadvisor, Uber, and Spotify. Anthropic also improved how connectors surface in conversation — Claude now suggests the right app in context, and users can search, plan, and act without leaving the chat.
200+ connectors now available. New consumer apps: AllTrails, Instacart, Audible, Tripadvisor, Uber, Spotify. Improved in-context connector suggestions — Claude surfaces the relevant app during conversation rather than requiring the user to navigate to it. Multi-app workflows in a single conversation are now a first-class pattern (example: pull Amplitude data → build Canva deck → drop link into Asana, all in one chat).
Directory: 200+ connectors | Categories: design, finance, productivity, health, entertainment, transport | In-context suggestion: Claude recommends connectors mid-conversation | MCP app for financial services (Moody's) also added this week
Best for: All Claude users — especially those wanting to combine multiple apps in one conversation
New enterprise AI services company — Blackstone, H&F, Goldman Sachs
Anthropic announced the formation of a new enterprise AI services company in partnership with Blackstone, Hellman & Friedman, and Goldman Sachs. The company will provide AI deployment, integration, and managed services for large enterprise customers, with Claude at the center of the offering. This is separate from Anthropic's direct enterprise product sales.
New enterprise AI services company announced with Blackstone, Hellman & Friedman, and Goldman Sachs. Targets large enterprise AI deployment and managed services. Details on structure and launch timeline to follow.
Partners: Blackstone, Hellman & Friedman, Goldman Sachs | Focus: enterprise AI deployment and managed services | Distinct from Anthropic's direct Enterprise plan sales
Best for: Large enterprises evaluating managed AI deployment with financial-services backing
Finance agent templates + Microsoft 365 add-ins coming
Anthropic released ten ready-to-run agent templates for financial services workflows this week — covering pitchbook building, KYC file screening, and month-end close. A Moody's MCP app brings Claude into financial data workflows. Microsoft 365 add-ins for Excel, PowerPoint, Word, and Outlook are also confirmed as coming soon.
Ten finance agent templates released: pitchbooks, KYC screening, month-end close, and more. Moody's MCP app available for financial services and insurance. Microsoft 365 Word and Outlook add-ins announced (Excel and PowerPoint add-ins already live) — timeline to be confirmed.
Finance templates: 10 ready-to-run | Moody's MCP: financial services and insurance workflows | M365: Excel + PowerPoint already live; Word + Outlook announced coming soon
Best for: Financial services teams, investment banks, insurance firms, enterprise finance operations
Claude Code — stability and reliability fixes
Alongside the major feature announcements, Claude Code shipped several point releases this week focused on stability, rendering, and session handling.
Large repo optimization: startup memory reduced by ~80MB on 250,000-file repositories. Fixed: sessions on 1M-context models being falsely blocked with 'Prompt is too long' before reaching the actual API limit. Fixed: parallel shell tool calls where a failing read-only command (grep, git diff, ls) cancelled sibling calls. Fixed: VS Code extension failing to activate on Windows due to a hardcoded build path. Fixed: Mantle endpoint authentication failing with missing x-api-key header. Fixed: markdown link labels being lost on terminals without OSC 8 hyperlink support. Fixed: Bedrock and Vertex 400 errors when ENABLE_PROMPT_CACHING_1H is set. MCP connectors: unauthorized claude.ai connectors now correctly show 'needs auth' instead of 'failed'.
Memory: ~80MB saved on 250k-file repos | 1M context false-positive block fixed | Parallel shell fix: failing grep/git diff/ls no longer cancels siblings | VS Code Windows fix: createRequire polyfill bug | Mantle fix: x-api-key header | OSC 8 fallback: links now render as label (url) | ENABLE_PROMPT_CACHING_1H: Bedrock/Vertex 400 error fixed
Best for: Developers on large codebases, Windows users, Bedrock/Mantle users, VS Code users
Plans and Pricing
No pricing changes this week. The main news for subscribers is the effective limit increase: Claude Code five-hour limits doubled and peak-hour reductions removed for Pro and Max, effective immediately as of May 6.
Opus 4.7: $5/$25 per MTok | Sonnet 4.6: $3/$15 per MTok | Haiku 4.5: low-cost tier | Claude Code limits: 2x five-hour limit for all paid tiers; peak-hour cap removed for Pro + Max | Managed Agents: contact enterprise sales
Best for: No action needed on pricing — limits upgrade is automatic for all paid subscribers
ChatGPT / OpenAI
Dateline: May 08, 2026 | Next update: May 15, 2026
Over the past week, OpenAI has focused on refining system behaviour in professional contexts, improving agent reliability, and continuing incremental improvements to multimodal and memory performance.
GPT-5.3 Standard — default model
More consistent performance across longer and more complex interactions.
Improved handling of multi-turn conversations with reduced drift.
Context ~128k | Output ~4k–8k | Improved conversational state tracking
Best for: General use
GPT-5.3 Pro — high-reasoning model
Better stability in extended analytical and structured tasks.
Improved logical consistency across long outputs.
Context ~200k (est.) | Reduced reasoning inconsistencies over long chains
Best for: Deep analysis
GPT-5.3 Mini — fallback model
More seamless integration in mixed-model responses.
Less noticeable transitions when fallback occurs mid-task.
Context ~64k | Improved response blending
Best for: Quick tasks
Agent Mode
Handles multi-step workflows.
Improved ability to stay aligned with original task objectives over longer sequences.
Reduced task drift | Better instruction persistence
Best for: Task delegation
Deep Research
Combines browsing and reasoning.
More structured outputs when synthesising complex material.
Improved formatting layer on top of synthesis pipeline
Best for: Research
Memory & Projects
Persistent context across chats.
Improved filtering of outdated or low-relevance memories.
Better decay mechanisms in memory prioritisation
Best for: Ongoing workflows
Advanced Voice Mode
Improved conversational flow.
More natural conversational pacing in longer exchanges.
Improved timing and response smoothing
Best for: Voice interaction
ChatGPT for Clinicians
Ongoing refinement of healthcare-focused workflows.
Improved clarity and structure in generated clinical documentation outputs.
Further tuning of domain-specific output formats | Strengthened safety constraints
Best for: Clinical support (non-diagnostic assistance)
Other Features & Pricing
Continued improvements in multimodal consistency (text, images, documents). General stability improvements across integrations and tools. No major updates to Sora, Codex, or pricing.
What this means
This week is another quiet but strategically important refinement cycle.
Three things matter:
First, the continued improvements to multi-turn consistency and task persistence show that OpenAI is prioritising a core weakness of current systems: losing track of context. Fixing this is essential if AI is to move from a "tool you prompt" to a system you rely on over time.
Second, Agent Mode is clearly converging toward longer-horizon autonomy. The improvements are not dramatic, but they are directional — less drift, better alignment, more reliable execution. That is exactly what is needed before any meaningful scale-up of autonomous systems.
Third, the ongoing iteration on ChatGPT for clinicians reinforces a broader pattern: OpenAI is doubling down on professional verticals, not just releasing general tools. Healthcare is being treated as a test case for how AI can enter regulated, high-stakes environments.
Overall, this is a stability and alignment week. The technology is not jumping forward in capability, but it is becoming more predictable, controllable, and usable in real-world settings — which, at this stage, matters more.
Gemini
Date: May 8, 2026 | Next update: May 15, 2026
As we approach Google I/O 2026 (scheduled for May 19), this week's focus has been on hardware integration for the home and a significant release for developers: the General Availability of Gemini 3.1 Flash-Lite.
Gemini 3.1 Flash-Lite — now GA
Officially reached General Availability on May 7, 2026. This version is fully optimized for maximum speed and cost efficiency for developers building high-scale applications.
Best for: High-volume apps, high-scale developer applications
Gemini 2.5 Flash update
A newly enhanced version of the 2.5 Flash model was rolled out to the Gemini app. Features significantly improved image understanding and better formatting (headers, tables, and lists) for organizing complex information.
Best for: Learning & organization, enhanced table formatting
Gemini for Home — major upgrade
Google Home voice assistants received a "Gemini 3.1 intelligence upgrade." This allows users to issue complex, multi-step voice commands in a single breath (e.g., "Lock the doors, turn off the lights, and set the alarm").
Multi-step voice commands now supported via Gemini 3.1 on Google Home devices.
Best for: Smart home control
"Ask Home" web interface preview
Google previewed a new conversational web interface for home.google.com. It will allow users to search camera history and manage smart devices via natural language from any browser.
Switching tools & AI Memories
New tools launched in Settings allow users to import their chat history and personal context from other AI apps into Gemini. Additionally, "past chats" are being rebranded as "Memories."
New "Import" tools for AI memories and chat history — bring context from other AI apps into Gemini.
Workspace "Persistent Instructions"
An update to Gemini in Google Docs now supports persistent instructions, allowing the AI to remember specific formatting or style preferences throughout a document.
DeepMind & gaming research
Google DeepMind announced a research partnership with Fenris Creations (formerly CCP Games). They will use an offline version of EVE Online to study AI memory and long-horizon planning in complex systems.
Plans and upcoming events
Google I/O 2026: The official schedule is now live. The main keynote is set for May 19 at 10:00 AM PT, where major model and Android integrations are expected.
"Google Health" transition: It was announced that the Fitbit app will be replaced by Google Health later this year. A new "Premium" plan will join the AI Pro subscription tier.
Service status: Minor streaming issues with Deep Research were detected on May 3 and May 5, but both were resolved within hours. API key serving is currently stable following some intermittent latency in late April.
Microsoft Copilot
Dateline: May 8, 2026 | Next update: May 15, 2026
The biggest story of the week is integration: Microsoft expanded Copilot's reach across productivity and developer ecosystems, introducing deeper Office 365 add-ins, new workflow automation templates, and a major boost in GitHub Copilot X capabilities. Also this week: Copilot Studio gained multi-agent orchestration, connectors expanded to over 150 apps, and a new enterprise AI deployment program was announced with Accenture and Deloitte.
Microsoft 365 integration — Word & Outlook add-ins live
Copilot add-ins for Word and Outlook are now live, completing the Office suite integration (Excel and PowerPoint were already available). Users can now draft, summarize, and auto-respond directly inside Outlook, and generate structured documents in Word with AI assistance.
Word + Outlook add-ins released. Excel + PowerPoint already live. Enterprise rollout begins this week.
Office 365 add-ins: Word, Outlook, Excel, PowerPoint | Context-aware drafting and summarization | Admin controls via Microsoft 365 compliance center
Best for: Enterprise and business users managing documents and communications at scale
GitHub Copilot X — expanded developer features
GitHub Copilot X introduced multi-file context awareness and CI/CD auto-fix this week. Developers can now rely on Copilot to review pull requests across multiple repositories and automatically suggest fixes.
Multi-file context support. CI/CD auto-fix for pipeline errors. Expanded API usage limits for enterprise teams.
Multi-file context: up to 500k tokens across repos | CI/CD auto-fix: automated PR fixes | API limits: doubled for enterprise GitHub Copilot X subscribers
Best for: Developers, engineering leaders, enterprise DevOps teams
Copilot Studio — multi-agent orchestration
Copilot Studio gained multi-agent orchestration, allowing a lead agent to break tasks into subtasks and delegate them to specialized agents. This mirrors Anthropic's orchestration approach but is tightly integrated with Microsoft Teams and Azure.
Multi-agent orchestration with shared workspace. Full traceability in Copilot Console.
Lead agent + parallel subagents | Shared workspace in Teams | Trace logs in Copilot Console | Configurable autonomy levels
Best for: Enterprise teams building complex workflows, investigative research, and document-heavy projects
Connectors — now 150+ apps including LinkedIn, Dropbox, Eventbrite
The Copilot connector directory crossed 150 apps this week, adding LinkedIn, Dropbox, and Eventbrite. Improved in-context suggestions now surface the right connector mid-conversation.
150+ connectors available. LinkedIn, Dropbox, Eventbrite added. Multi-app workflows supported in a single chat.
Directory: 150+ connectors | Categories: productivity, social, storage, events | In-context suggestion: Copilot recommends connectors mid-conversation
Best for: All Copilot users — especially those combining multiple apps in one workflow
Enterprise AI deployment program — Accenture & Deloitte partnership
Microsoft announced a new enterprise AI deployment program in partnership with Accenture and Deloitte, focused on managed services, integration, and compliance for large organizations.
Enterprise AI deployment program launched with Accenture and Deloitte. Targets Fortune 500 adoption.
Partners: Accenture, Deloitte | Focus: enterprise AI deployment, compliance, managed services | Distinct from direct Microsoft 365 sales
Best for: Large enterprises evaluating managed AI deployment
Workflow templates — finance & HR automation
Ten new workflow templates were released for finance and HR, covering payroll reconciliation, compliance checks, and onboarding automation.
Finance + HR templates released. Payroll reconciliation, compliance, onboarding automation.
Templates: 10 ready-to-run | Integrated with Teams + Power Automate | HR + finance workflows
Best for: Finance teams, HR departments, enterprise operations
Copilot reliability & fixes
Large document optimization: memory reduced by ~60MB on 100k-word files. Fixed: Teams connector failing on multi-tenant accounts. Fixed: GitHub Copilot X extension crash on macOS. Fixed: Outlook add-in authentication loop.
Memory: ~60MB saved | Teams multi-tenant fix | GitHub Copilot X macOS crash resolved | Outlook add-in auth loop fixed
Best for: Heavy document users, Teams admins, GitHub developers
Plans and Pricing
No pricing changes this week. The main news for subscribers is the effective limit increase: GitHub Copilot X API limits doubled for enterprise teams. All upgrades are automatic for existing subscribers.
GitHub Copilot X: doubled API limits | Office 365 add-ins: included in Enterprise plan | Copilot Studio orchestration: enterprise only