Dateline: June 12, 2026 | Next update: June 19, 2026
The biggest Anthropic launch of 2026: on June 9, Claude Fable 5 and Claude Mythos 5 arrived simultaneously, introducing an entirely new model tier above Opus. Fable 5 is the publicly available Mythos-class model — the most capable Claude ever released for general use — while Mythos 5 is the same underlying model with safeguards lifted, restricted to Project Glasswing partners. Alongside the model launch, Managed Agents gained scheduled deployments and credential vaults; the Claude Platform added connector observability and MCP tunnel support in the Messages API; Enterprise got custom admin roles; and Claude Code shipped nested sub-agents, a post-session hook, and a batch of stability fixes.
Claude / Anthropic
Claude Fable 5 — first publicly available Mythos-class model
Claude Fable 5 is Anthropic's most capable publicly available model by a significant margin. It is the first member of the Mythos-class tier — a new capability level above Opus — made safe for general release. Anthropic describes it as state-of-the-art on nearly every benchmark tested, with its largest performance advantage on long, complex, and autonomous tasks. Stripe completed a 50-million-line Ruby codebase migration in one day that had previously been estimated at over two months for a human team. The model can work for days at a time in agentic harnesses, planning across stages, delegating to sub-agents, and verifying its own work.
Claude Fable 5 (claude-fable-5) launches June 9 as the first publicly available Mythos-class model. Key capabilities: agentic coding (SWE-Bench Pro: ~80%, per independent testing); always-on adaptive thinking; state-of-the-art vision (extracts numbers from scientific figures, rebuilds web apps from screenshots, completed Pokémon FireRed using only vision); long-context performance (with file-based memory, scored 3x better than Opus 4.8 on Slay the Spire); and senior-researcher-level knowledge work across finance, legal, analytics, and architecture. Safety classifiers: Fable 5 and Mythos 5 share the same underlying weights. Fable 5 ships with safety classifiers that silently reroute sensitive queries (cybersecurity, biology, chemistry) to Claude Opus 4.8 as a fallback — fewer than 5% of sessions are affected. Rerouted responses return stop_reason: 'refusal' as an HTTP 200 and are not billed if no output is generated. Claude Mythos 5 (claude-mythos-5) launches simultaneously with safeguards lifted in specific areas, available only to Project Glasswing partners and trusted-access enterprise users. Both models carry a 30-day data retention requirement and are not available under zero data retention. Subscription plans: Fable 5 is included free on Pro, Max, Team, and seat-based Enterprise from June 9–22. After June 22, usage credits will be required until capacity allows it to return as a standard plan feature.
Model strings: claude-fable-5, claude-mythos-5 | Context: 1M tokens (API, Bedrock, Vertex AI); 200k (Microsoft Foundry) | Max output: 128k tokens | Pricing: $10/$50 per MTok standard; $5/$25 per MTok Batch API | US-only inference: 1.1x pricing for data residency | Subscription: 2x usage credit count on claude.ai | Data retention: 30 days (Covered Model, ZDR not available) | GitHub Copilot: admin policy off by default, 30-day retention required | Bedrock: available on bedrock-runtime; bedrock-mantle console support coming soon | Update to Claude Code v2.1.170 for Fable 5 access
Best for: Long-horizon agentic coding, large-scale migrations, complex knowledge work, multi-stage research — the longer and more complex the task, the larger Fable 5's advantage
Claude Managed Agents — scheduled deployments and credential vaults
Two features that have been blocking enterprise agent projects from production deployment shipped together on June 9: agents can now run on a cron schedule without any human trigger, and sensitive credentials can be stored in platform-managed vaults rather than being embedded in prompts or passed as environment variables in plaintext. Rakuten is already using scheduled deployments to produce weekly and monthly spreadsheet analysis reports automatically. Milana uses credential vaults to connect its AI product engineer to customer codebases for automated bug detection.
Scheduled deployments (public beta): set agents to run on a cron schedule — hourly, daily, weekly, or custom — with no human trigger required. Agents receive a start message, complete their work, and log outputs automatically. Vault environment variables (public beta): store API keys, tokens, and other credentials in a platform-managed vault; the agent retrieves them at runtime without those values ever appearing in prompts, logs, or conversation history. Both features are managed from the Claude Console. Richer session thread webhook events also added — subscribe to lifecycle events (agent started, tool call made, result produced, agent finished) for integration with external monitoring and orchestration systems.
Scheduled deployments: cron-based, configured in Claude Console | Vault credentials: stored encrypted, retrieved at runtime, never exposed in conversation | Webhook events: session thread lifecycle (start, tool call, result, finish) | Both in public beta under managed-agents-2026-04-01 header | MCP tunnels: now also supported in the Messages API (not only Managed Agents); managed from Claude Console by org admins
Best for: Enterprise teams running recurring agent workflows (reporting, analysis, monitoring) and any team storing credentials for agent tool access
Claude Platform — connector observability and MCP tunnels in Messages API
Two platform-level updates shipped alongside the Fable 5 launch. Connector observability gives workspace admins visibility into how MCP connectors are being used across all Claude products. MCP tunnels — previously available only in Managed Agents — are now also supported directly in the Messages API, opening private service access to any Claude API integration.
Connector observability (public beta): admins and owners can now monitor adoption rates, error rates, latency, and usage volumes for every MCP connector deployed across their organisation's Claude products. In-app directory submission also added — submit custom MCP connectors to the Claude connector directory directly from within Claude, without going through a separate web form. MCP tunnels in Messages API: the encrypted tunnel mechanism for reaching private databases, internal APIs, and authenticated services is now available in the Messages API (not only Managed Agents). Managed from the Claude Console by organisation admins.
Connector observability: metrics include adoption, errors, latency, usage volume | In-app submission: submit to Claude connector directory from within Claude | MCP tunnels: now in Messages API; managed via Claude Console | Complements: self-hosted sandboxes (Managed Agents public beta) and MCP tunnels in Managed Agents (research preview) announced May 19
Best for: Enterprise admins monitoring connector health; API developers needing access to private internal services without public endpoints
Claude Enterprise — custom admin roles with granular permissions
Enterprise plans can now configure custom roles with granular admin permissions. Previously, accessing billing or privacy settings required full Owner access. Members can now be granted only the specific admin permissions they need, without elevating them to Owner.
Custom roles now support admin permissions on Enterprise plans. Assign billing access, privacy settings access, or other admin functions to specific members without giving them full Owner access. Built on the existing custom roles framework — no new configuration system to learn.
Feature: admin permission granularity in custom roles | Plans: Enterprise | Manages: billing access, privacy settings, and other admin functions independently of Owner role | Complements: SCIM/RBAC and existing custom roles already available on Enterprise
Best for: Enterprise IT admins and workspace owners who need to delegate specific admin tasks without granting full ownership
Claude Code — Fable 5 support, nested sub-agents, post-session hook
Claude Code's update this week is anchored by Fable 5 integration (update to v2.1.170), but also ships nested sub-agent support, a PostSession hook for cleanup and logging, a safe mode for troubleshooting, and a /cd command for moving a session's working directory mid-conversation.
Fable 5 available in Claude Code from v2.1.170. Nested sub-agents: sub-agents can now spawn their own sub-agents, enabling deeper parallel agent trees for complex Dynamic Workflows. PostSession hook: runs after a session ends, enabling automated cleanup, logging, or notification actions. New /cd command: change the working directory of the current session without restarting. Safe mode (--safe-mode flag): disables plugins and MCP servers for troubleshooting. New /plugin search command: search the marketplace from within Claude Code without opening a browser. Smarter model and region handling: Claude Code now selects the optimal inference region automatically for Fable 5 based on availability and latency.
Update required: v2.1.170 for Fable 5 | Nested sub-agents: sub-agents can spawn sub-agents (Dynamic Workflows research preview) | PostSession hook: runs after session close | /cd: change working directory in-session | --safe-mode: disables plugins and MCP | /plugin search: in-terminal marketplace search | Fixed: VS Code integrated terminal transcript bug | Fixed: claude -p hanging when a backgrounded command never exits (background shells stopped ~5s after result once stdin closes) | Fixed: claude -p failing with ANTHROPIC_API_KEY required on Bedrock/Vertex/Foundry when CI=true | Fixed: Bash commands failing under bazel and EDR-protected Go workflows ($TMPDIR regression from 2.1.154) | Fixed: Windows EEXIST error on session-env directory in OneDrive or read-only paths | Fixed: org-managed permission rules not applying for the entire session when managed settings fetch completed during startup on a fresh config directory
Best for: All Claude Code users — update to v2.1.170 immediately for Fable 5 access and the bazel/$TMPDIR regression fix
⚠️ Model deprecations — Sonnet 4 and Opus 4 retire June 15
Claude Sonnet 4 and the original Claude Opus 4 retire on June 15 — three days from today. Requests to these model strings will begin failing after that date. Migrate to Sonnet 4.6 and Opus 4.8 (or Fable 5) respectively.
June 15 retirement is three days away. Sonnet 4 → migrate to claude-sonnet-4-6. Original Opus 4 → migrate to claude-opus-4-8 or claude-fable-5. Check all integrations, CI pipelines, and hardcoded model strings now.
Retiring June 15: claude-sonnet-4-20250514, claude-opus-4-20250514 | Migration: Sonnet 4 → claude-sonnet-4-6 | Opus 4 → claude-opus-4-8 or claude-fable-5 | Already retired (April 19): claude-3-haiku-20240307
Best for: API developers — immediate action required
Plans and Pricing
Fable 5 is the first meaningful pricing change in several weeks. At $10/$50 per MTok it is priced at 2x Opus 4.8, but more than 50% cheaper than the Mythos Preview rate that Glasswing partners had been paying. Batch API pricing for Fable 5 is $5/$25 per MTok — the same as Opus 4.8 standard. On subscription plans, Fable 5 is free until June 22, after which usage credits will be required until capacity scales.
Fable 5 standard: $10/$50 per MTok | Fable 5 Batch API: $5/$25 per MTok | Fable 5 US-only inference: 1.1x pricing | Subscription (Pro/Max/Team/Enterprise): free June 9–22; usage credits required from June 23 until capacity restored; counts as 2x usage | Opus 4.8: $5/$25 per MTok (unchanged) | Sonnet 4.6: $3/$15 per MTok | Haiku 4.5: low-cost tier
Best for: API users: use Batch API at $5/$25 for non-latency-sensitive Fable 5 workloads. Subscription users: maximise Fable 5 use before June 22 while it is still free on all paid plans.
ChatGPT / OpenAI
Dateline: June 12, 2026 | Next update: June 19, 2026
Over the past week, OpenAI has focused on expanding ChatGPT's enterprise capabilities, improving agent performance, and strengthening the platform's position as a persistent workspace for research and knowledge work.
GPT-5.3 Standard — default model
Improved context retention when switching between research, drafting, and analysis tasks within a single session.
Context ~128k | Output ~4k–8k | Improved cross-task context management
Best for: General use
GPT-5.3 Pro — high-reasoning model
Improved handling of large research projects requiring multiple rounds of revision and synthesis.
Context ~200k (est.) | Improved long-horizon reasoning consistency
Best for: Deep analysis
GPT-5.3 Mini — fallback model
Improved routing efficiency, reducing latency for routine requests.
Context ~64k | Better inference allocation and task classification
Best for: Quick tasks
Agent Mode
Improved coordination between planning, execution, and verification stages of complex tasks.
Enhanced workflow orchestration | Better state persistence across task stages
Best for: Task delegation
Deep Research
Improved source evaluation and evidence synthesis across large sets of material.
Enhanced ranking and conflict-resolution systems | Better citation selection
Best for: Research
Memory & Projects
Improved retrieval of project-specific information while reducing interference from unrelated conversations.
Better project isolation | More precise memory retrieval
Best for: Ongoing workflows
Advanced Voice Mode
More natural handling of topic transitions and follow-up questions.
Improved conversational state management | Reduced context loss during transitions
Best for: Voice interaction
Enterprise & Team Collaboration
Improved support for collaborative projects, shared research workflows, and team-based knowledge management.
Enhanced collaboration architecture | Improved project coordination systems
Best for: Organisational productivity
Professional Workflows
Better structured outputs for reports, planning documents, policy briefs, and research summaries.
Improved document-generation consistency | Better workflow-specific optimisation
Best for: Research and professional writing
Plans and Pricing
No significant pricing changes announced this week. Pricing stable | API structure unchanged.
Gemini (Google)
Date: June 12, 2026 | Next update: June 19, 2026
This week's updates mark the final enforcement of several core API deprecations alongside critical compliance additions for enterprise administration. The week was also highlighted by a widespread global service disruption that temporarily impacted Workspace integrations.
Gemini 2.0 series — permanently retired
The legacy Gemini 2.0 models (gemini-2.0-flash, gemini-2.0-flash-001, gemini-2.0-flash-lite, and gemini-2.0-flash-lite-001) have been completely shut down. Developers must permanently transition active endpoints to either gemini-3.5-flash or gemini-3.1-flash-lite.
All Gemini 2.0 model endpoints are offline as of June 1, 2026. Any remaining traffic to gemini-2.0-flash, gemini-2.0-flash-001, gemini-2.0-flash-lite, or gemini-2.0-flash-lite-001 will fail. Migrate to gemini-3.5-flash or gemini-3.1-flash-lite immediately.
Best for: All developers — check every integration for Gemini 2.0 strings now
Image model deprecations — June 25 deadline
Both gemini-3.1-flash-image-preview and gemini-3-pro-image-preview are officially scheduled to be turned off on June 25, 2026. Migrate image generation workflows before that date.
Best for: Developers using image preview models — migrate before June 25
Google Vault — new admin compliance for Gemini
Google Vault now natively supports retention rules and litigation holds for the Gemini app across web and mobile platforms. Administrators can now actively enforce compliance policies, extending previous capabilities that only permitted searching and exporting chat results.
Retention rules and litigation holds for Gemini chat history now available in Google Vault. Admins can enforce compliance policies on Gemini content across web and mobile — previously limited to search and export only.
Best for: Enterprise admins, legal and compliance teams, regulated industries
Gemini in Google Classroom — image-to-rubric
Educators can now upload .jpeg and .png images and use Gemini to directly convert them into organised Google Classroom rubrics during assignment setup.
Best for: Educators and academic institutions
Ask Gemini in Drive — Gmail integration now GA
The integration of Gmail as an active knowledge source inside Ask Gemini in Drive has officially moved out of beta and into General Availability.
Best for: Knowledge workers using Drive and Gmail together
Workspace — AI Plus price cut
Google adjusted its consumer tier structure, cutting the AI Plus subscription down to $4.99 while simultaneously doubling the associated storage limits to 400 GB.
Best for: Individual users considering the AI Plus tier
API schema cutoff — enforced
The mandatory breaking change to the Interactions API structure was completed on June 8. The legacy formatting structure has been removed, and the API now strictly defaults to the new agentic workflow schema (outputs → steps). Any code still reading the outputs field will fail.
Best for: Agent developers — audit all code referencing the outputs field immediately
Google Calendar API — writerWithoutPrivateAccess coming June 29
The new writerWithoutPrivateAccess permission rule in the Calendar API becomes generally available on June 29, 2026. Developers should prepare integrations now.
Best for: Developers with Calendar API integrations — review permissions before June 29
Global Workspace outage — June 10
On June 10, the Gemini app and Workspace side panel experienced a major multi-hour service disruption. Standard web and mobile platforms, alongside Gemini in Chrome, returned widespread errors (codes 1076 and 1099). Engineering teams applied server-side mitigations, resolving the issue completely by 10:30 AM PDT.
Best for: No action needed — fully resolved. Monitor the Google Workspace Status Dashboard for future incidents.
Microsoft Copilot
Dateline: June 12, 2026 | Next update: June 19, 2026
The biggest story of the week is operational scale and compliance: Copilot is shifting from feature additions to enterprise-ready governance, agent extensibility, and structured data integration.
SharePoint list support in Agent Builder
Agents can now use SharePoint lists as structured data sources (up to 20,000 items). Unlocks inventory tracking, HR requests, task boards, and other operational data use cases — moving Copilot beyond document-based retrieval into live structured data.
Structured list integration | Up to 20,000 items per list | Declarative agent access | Moves beyond documents into operational data
Best for: Enterprises managing structured workflows — HR requests, inventory, project boards
Vision inputs in Copilot Chat
Copilot Chat now supports vision inputs, allowing users to upload images for analysis, summaries, or contextual grounding. Supports diagrams, charts, and screenshots. Integrated with Work IQ for organisation-aware responses.
Image recognition + contextual Q&A | Supports diagrams, charts, screenshots | Integrated with Work IQ
Best for: Analysts, designers, educators, technical teams
Dynamic tool discovery for agents
Agents can now discover and load tools dynamically based on context, reducing manual configuration. Unified REST endpoint via Work IQ with declarative agent access and remote MCP server support.
Unified REST endpoint via Work IQ | Declarative agent access | Remote MCP server support
Best for: Enterprise developers building extensible agent workflows
Purview compliance — tighter governance
Expanded sensitivity labels, DLP signals, and eDiscovery coverage for Copilot and Loop content. Endpoint DLP health dashboard, insider risk policy recommendations, and connected experiences blocking via sensitivity labels.
Endpoint DLP health dashboard | Insider risk policy recommendations | Connected experiences blocking via sensitivity labels
Best for: Regulated industries, compliance officers, IT admins
Teams — meeting templates and AI audio summaries
Teams now offers meeting templates for recurring workflows and AI audio summaries for news and events, integrated into Teams feeds.
Event templates for recurring workflows | AI audio summaries integrated into Teams feeds
Best for: Meeting-heavy organisations, communications teams
Outlook — inbox rules and rewrite improvements
Outlook Copilot now supports external tag detection in inbox rules, preview line customisation, and improved email rewrite controls for clarity and tone.
Preview line customisation | External tag detection | Rewrite improvements for clarity and tone
Best for: Business users, compliance-sensitive organisations
PowerPoint — brand kits and custom skills
Copilot in PowerPoint now supports brand kits and custom skills for presentations. Select brand menu, skills for layout and content, integrated with Copilot Notebooks for deck creation.
Select brand menu | Skills for layout/content | Integration with Copilot Notebooks
Best for: Marketing teams, enterprise communications, educators
Admin checklist — June 2026
Microsoft recommends five admin actions this month:
- Establish Copilot-ready data sources before scaling agents
- Re-evaluate security and compliance posture with expanded Purview coverage
- Define agent governance before broad deployment
- Pilot high-impact use cases first
- Prepare support and end users with updated training
Plans and Pricing
No pricing changes this week. Updates focus on agent extensibility, compliance, and structured data integration. SharePoint list support included in Agent Builder; vision inputs in Copilot Chat; compliance updates in Purview; brand kits in PowerPoint enterprise tier.