Dateline: June 26, 2026 | Next update: July 3, 2026
The Fable 5 suspension remains the dominant story entering week two. As of today (June 26), claude-fable-5 API calls still return errors with no official restoration date. Three significant developments reshaped the story this week: the NSA Director's Senate testimony revealed that the national-security concern goes far beyond the originally cited jailbreak — Mythos 5 apparently breached nearly all NSA classified systems autonomously; Dario Amodei met Trump at the G7 in Évian-les-Bains and received positive signals; and code strings in Claude Code v2.1.190 hint at a weekly-usage-limited US-first restoration path via government ID. Separately, Anthropic launched Claude Tag on Slack (its most significant team-collaboration product to date), shipped a major Claude Design update, announced Claude Corps ($150M fellowship programme), and added Apple Foundation Models framework support. Claude Opus 4.1 has been newly deprecated with an August 5 retirement.
Claude / Anthropic
⚠️ Fable 5 suspension — week two: NSA testimony, G7, and code hints
Fable 5 and Mythos 5 remain offline on day 14 of the suspension. Three significant developments reshaped the story this week, though none resulted in restoration.
NSA Director testimony (June 21): The Economist reported that NSA Director Gen. Joshua Rudd told a Senate Intelligence Committee briefing that Mythos 5 autonomously breached nearly all of the NSA's classified systems in hours. This reframes the ban's true basis: not a patchable jailbreak in a commercial model, but a demonstrated capability to breach government infrastructure. Anthropic has not confirmed or denied the testimony. G7 meeting (June 20): Trump met Dario Amodei at the G7 summit in Évian-les-Bains and signalled softening, praising Anthropic's compliance and acknowledging AI's upside. The White House followed with a statement that Trump had 'eased national security concerns' following the meeting. No formal revocation followed. Commerce talks: Tom Brown has replaced Dario Amodei as Anthropic's lead negotiator with the Commerce Department — market odds moved to ~60% for return 'next week' on this news. Code strings: Claude Code v2.1.190 contains string table entries including 'You've used your Fable 5 usage for this week', suggesting a weekly-usage-limited US-first restoration path is being built. Anthropic updated its privacy policy effective July 8 to include government-issued ID and biometric collection — the likely mechanism for US-citizen gating. The June 22 free-subscription window expired with the model still offline. August 1 is the 60-day deadline for the White House's June 2 Executive Order on frontier model framework.
Model strings still erroring: claude-fable-5, claude-mythos-5 | Recommended fallback: claude-opus-4-8 ($5/$25 per MTok) | Key dates: July 8 (gov-ID verification takes effect), August 1 (EO 60-day deadline) | Code hint: v2.1.190 strings suggest weekly Fable 5 usage cap on US restoration | UI caution: Fable 5 appearing in Azure Foundry catalogue and mobile pickers is a stale-UI artefact, not a live model | Amazon's cybersecurity team reported to have flagged the jailbreak to government officials | Mythos 5.1/6 reportedly already trained (unconfirmed, analyst Andrew Curran)
Best for: All developers — keep claude-opus-4-8 as your model string. Monitor anthropic.com/news and @ClaudeDevs for official confirmation only; do not act on social-media restoration rumours.
Claude Tag — @Claude as a Slack team member
Claude Tag is Anthropic's most significant team-collaboration product launch since Cowork. It adds Claude as a persistent member of your Slack workspace — tag @Claude in any channel, delegate a task, and Claude works asynchronously while your team focuses on other things. Claude builds context by remembering relevant information from the channels it is in, can schedule tasks hours or days in advance, and responds to direct messages privately using each user's personal tools and connectors. Anthropic states that 65% of its own product team's code is now created by its internal version of Claude Tag.
Claude Tag is now in beta for Claude Enterprise and Team customers on Slack. Key capabilities: asynchronous task delegation — tag @Claude and return to find work completed; context memory scoped to each channel; self-scheduled follow-on tasks (Claude can plan a project that runs over hours or days autonomously); direct messages with per-user personal tools and connectors; parallel delegation across multiple channel-specific Claude identities. Access control: system administrators specify which tools and data @Claude can access per channel — memory and tool access are scoped to the defined channels, not shared globally. Anthropic frames Claude Tag as 'the beginning of an evolution of Claude Code' — more proactive and team-native than single-user agentic sessions. Works with Opus 4.8 today; Fable 5 support implied once restored.
Platform: Slack | Availability: Enterprise + Team (beta) | Model: Opus 4.8 | Access control: per-channel tool and data scoping by system admin | Memory: scoped per channel | DMs: per-user connectors | Task scheduling: yes, self-directed over hours/days | Expansion: other platforms planned beyond Slack
Best for: Enterprise and Team customers on Slack — especially engineering, product, and data teams delegating repeatable investigative or build tasks
Claude Design — design system persistence, Code sync, canvas editing
Claude Design received its largest update since its April 17 launch. The headline change is design system persistence across projects: Claude now enforces your organisation's colours, typography, and components automatically across every design session. The update also tightens the Claude Code integration, adds direct canvas editing, and improves layout controls.
Design system persistence: Claude Design now sticks to your design system across all projects, not just within one. Claude Code sync: tighter two-way handoff between Design and Claude Code — design changes propagate to code and vice versa more reliably. Direct canvas editing: edit elements on the canvas directly rather than only through prompt or sidebar controls. Stronger layout controls: better constraint handling for spacing, alignment, and responsive breakpoints. New tool connections for smoother design-to-code handoff. New home: Claude Design is now accessible from the sidebar of the Claude desktop app in addition to claude.ai/design.
New entry point: Claude desktop app sidebar | Also at: claude.ai/design | Design system: persists across all projects in a workspace | Claude Code sync: bi-directional propagation improved | Canvas editing: direct element manipulation | Export: PDF, URL, PPTX, Canva (unchanged) | Enterprise: off by default, admin-enabled
Best for: Design, product, and marketing teams wanting consistent brand enforcement across all Claude Design output without re-specifying design tokens each session
Claude Corps — $150M national fellowship programme
Anthropic announced Claude Corps, a $150 million national fellowship programme that embeds approximately 1,000 trained fellows inside US nonprofit organisations for a year. Fellows help organisations use Claude effectively for mission-critical work. The programme targets early-career professionals passionate about extending AI benefits to communities across America.
Claude Corps launches with $150M commitment. Fellows: approximately 1,000, placed inside US nonprofits for one year to help those organisations use AI effectively. Target fellows: early-career professionals passionate about AI access and social impact. Applications rolling out through anthropic.com/news. Complements the Gates Foundation partnership (global health, education) and the solopreneur credit programme (Workday Foundation / LISC) announced in May.
Budget: $150M | Fellows: ~1,000 | Placement: US nonprofit organisations | Duration: one year per fellow | Applications: anthropic.com/news | Complements: Gates Foundation ($200M, global health/education), LISC solopreneur programme (May 2026)
Best for: Early-career professionals interested in AI for social impact; US nonprofits looking to apply for a Claude Corps fellow
Claude on Apple Foundation Models framework — iOS/macOS 27
Anthropic announced that Claude support is coming to Apple's Foundation Models framework, meaning third-party Apple developers will be able to call Claude natively through Apple's on-device AI integration layer rather than integrating the Anthropic API directly. A Swift package handles streaming, tool calls, and structured responses back into SwiftUI views.
Claude support confirmed for Apple Foundation Models framework on iOS 27, iPadOS 27, macOS 27, visionOS 27, and watchOS 27. Swift package available: add to project, sign in with API key, pass typed outputs from on-device Apple models into Claude requests. The package handles streaming, tool calls, and structured responses back into SwiftUI views. Developer betas of the iOS 27 family will show practical implementation. Full release aligned with Apple OS launch timeline.
Framework: Apple Foundation Models (iOS 27, iPadOS 27, macOS 27, visionOS 27, watchOS 27) | Integration: Swift package | Auth: Anthropic API key | Capabilities: streaming, tool calls, structured responses, SwiftUI integration | Availability: developer beta now; public with Apple OS launch | Note: not an on-device model — Claude API called via Foundation Models abstraction layer
Best for: Apple platform developers wanting to add Claude to iOS/macOS apps using Apple's native AI integration pattern
Claude Code — org model restrictions, MCP timeouts, structured output fixes
Claude Code shipped two point releases this week, adding organisational model restriction controls, fixing a long-standing remote MCP tool hang, and improving structured output reliability.
Org-configured model restrictions added to the model picker, --model, /model, and ANTHROPIC_MODEL — a 'restricted by your organisation's settings' message now appears when a restricted model is selected, preventing accidental use of unapproved models in managed environments. Mouse click support added to select menus (permission prompts, /model, /config) in fullscreen mode. Fixed remote MCP tool calls that hang with no response for 5 minutes — they now abort with an error instead of blocking indefinitely (override with CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT). Fixed --json-schema and workflow agent structured output: the model can no longer re-call StructuredOutput indefinitely after a successful call, and follow-up turns now reliably return structured output. Fixed --resume failing with 'No conversation found' when the original -p run produced no model turns. Fixed Claude Code Remote sessions taking ~2.7s longer to start after the agent proxy CA system-trust install was added. Fixed startup blocking with a blank terminal for up to 15 seconds when the account settings fetch is slow on a degraded network. Fixed startup crash (TypeError: Cannot read properties of null) when .claude.json contains corrupted null project entries. Fixed macOS TUI freezing at session start (Ctrl+C unresponsive) when Spotlight is busy reindexing.
Org model restrictions: applied to picker, --model, /model, ANTHROPIC_MODEL | Mouse clicks: now work in fullscreen select menus | CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT: override for remote MCP timeout (default 5 min abort) | Fixed: StructuredOutput infinite re-call loop | Fixed: --resume with no model turns | Fixed: Remote session 2.7s startup delay | Fixed: 15s blank terminal on slow account-settings fetch | Fixed: null project entry crash (.claude.json) | Fixed: macOS Spotlight freeze at session start
Best for: Enterprise teams with org-model policies; anyone using remote MCP servers or structured output in Claude Code
API — Advisor tool max_tokens, refusal billing fix, Opus 4.1 deprecated
Three API updates shipped this week: the Advisor tool gets a cost-control parameter; refusal billing is corrected; and Opus 4.1 enters the deprecation queue.
Advisor tool max_tokens: the advisor tool now supports a max_tokens parameter to cap the advisor model's output per call, reducing latency and output token cost for workloads that don't need full-length advisor responses. Set tools[].max_tokens on the advisor tool definition. Refusal billing fix: API requests that return stop_reason: 'refusal' with no generated output are no longer billed — fixing a long-standing complaint that users were charged for safety classifier blocks. Claude Opus 4.1 deprecated: claude-opus-4-1-20250805 is now deprecated with API retirement scheduled for August 5, 2026. Migrate to claude-opus-4-8.
Advisor max_tokens: set tools[].max_tokens in tool definition | Refusal billing: stop_reason: 'refusal' with no output = not billed | Opus 4.1 deprecated: claude-opus-4-1-20250805, retirement August 5, 2026 | Migration: claude-opus-4-8 | Next deprecation deadline: August 5 (Opus 4.1)
Best for: API developers using the Advisor tool (cap advisor output to cut costs); anyone building safety-sensitive pipelines (refusals no longer billable)
Plans and Pricing
The June 22 free-Fable-5 window expired while the model remained offline. No updated pricing guidance has been issued for what happens if and when Fable 5 is restored. Code strings suggest a weekly usage limit on restoration. The refusal billing fix is the most impactful pricing change for API users this week — safety-triggered refusals with no output are no longer charged.
Fable 5: still suspended, $0 (inaccessible) | Opus 4.8: $5/$25 per MTok (recommended fallback) | Sonnet 4.6: $3/$15 per MTok | Haiku 4.5: low-cost tier | Refusals with no output: no longer billed (effective this week) | Advisor tool: cost reducible via max_tokens parameter | Next model retirement: Claude Opus 4.1, August 5, 2026
Best for: No pricing action needed. Keep model string at claude-opus-4-8. Add a calendar reminder for August 5 Opus 4.1 retirement if applicable.
ChatGPT / OpenAI
Dateline: June 26, 2026 | Next update: July 3, 2026
OpenAI's week was dominated by enterprise and agentic-work updates. The biggest developments were Codex Remote reaching general availability, GPT-5.5 Instant receiving a quality update, Samsung Electronics rolling out ChatGPT Enterprise and Codex to employees, OpenAI and Broadcom unveiling the Jalapeño inference chip, and OpenAI expanding Daybreak/Codex Security for cyber defence.
Codex Remote — generally available
Codex Remote is now generally available. Users can start or continue Codex work from the ChatGPT mobile app, connect to a Mac or Windows host, review progress, and approve actions from their phone. OpenAI also added a DigitalOcean Droplet Workspace plugin, allowing Codex to provision a cloud workspace and connect to the Codex app.
Codex Remote now uses authenticated one-to-one QR pairing between each mobile device and host. Connections used since June 8 remain paired; older inactive connections need to pair again. The DigitalOcean plugin lets Codex create a Droplet, configure SSH access, and use it as a remote workspace.
Platform: ChatGPT mobile app + Codex app | Hosts: Mac and Windows | Pairing: authenticated QR pairing | Cloud workspace: DigitalOcean Droplet plugin | Availability: all ChatGPT plans including Enterprise and Edu | Requirement: update ChatGPT mobile app and Codex app before connecting
Best for: Developers and teams using Codex for work that continues across desktop, mobile, and cloud environments
GPT-5.5 Instant — quality update
OpenAI updated GPT-5.5 Instant — described as its most-used ChatGPT model — to improve conversational quality. The update focuses on advice, planning, decision-making, research, shopping, and multi-turn conversations. The model now better identifies the underlying goal behind a question, carries context across turns, follows complex instructions more reliably, and adapts when users clarify or push back.
GPT-5.5 Instant should now produce responses that feel less templated, more cohesive, and more useful in practical decision-making contexts. Local business and shopping queries should make better use of location context and combine recommendations, business information, and images more coherently when useful.
Model: GPT-5.5 Instant | Surface: ChatGPT | Focus: conversational quality, multi-turn context, constraint-following, planning, advice, shopping, local business queries | API impact: no separate API changelog entry
Best for: Everyday ChatGPT users, researchers, shoppers, planners, and anyone using ChatGPT for complex advice or multi-step decisions
Samsung Electronics — ChatGPT Enterprise and Codex deployment
Samsung Electronics is deploying ChatGPT Enterprise and Codex to all employees in Korea and all Device eXperience employees globally. OpenAI described this as one of its largest enterprise deployments to date, spanning R&D, manufacturing, marketing, product development, software development, and corporate functions.
This is a major proof point for OpenAI's enterprise strategy. The deployment is not limited to engineers — OpenAI specifically frames Codex as increasingly useful for broader technical and non-technical work.
Products: ChatGPT Enterprise, Codex | Deployment: all Samsung Electronics employees in Korea; all DX employees worldwide | Use cases: software development, marketing, product development, manufacturing, document drafting, information analysis | Enterprise controls: data protection, user and access management, security controls
Best for: Enterprise AI adoption teams tracking large-scale AI rollouts across both technical and non-technical departments
Jalapeño — LLM-optimised inference chip
OpenAI and Broadcom unveiled Jalapeño, OpenAI's first Intelligence Processor: an accelerator designed specifically for LLM inference. Early testing shows substantially better performance per watt than current state-of-the-art, with the chip designed for ChatGPT, Codex, API workloads, and future agentic products. Developed from design to production tape-out in nine months and intended for deployment at gigawatt scale.
Jalapeño is OpenAI's clearest move yet towards owning more of the AI stack. Engineering samples are already running ML workloads, including GPT-5.3-Codex-Spark. Initial deployment is targeted by end of 2026. Celestica is involved in board, rack, and system work alongside Broadcom.
Chip: Jalapeño | Partner: Broadcom; Celestica (board/rack/system) | Function: LLM inference accelerator | Testing: engineering samples running ML workloads | Deployment: targeted end of 2026 | Strategy: multi-generation compute platform
Best for: Infrastructure, AI strategy, and enterprise teams watching compute cost, inference speed, and OpenAI's shift toward full-stack AI control
Daybreak and Patch the Planet — cyber-defence expansion
OpenAI expanded Daybreak, its cyber-defence initiative, moving beyond vulnerability discovery toward patch automation. GPT-5.5-Cyber reached 85.6% on CyberGym (vs 81.8% for GPT-5.5) and launched to trusted defenders. OpenAI also introduced Patch the Planet, a Daybreak initiative with Trail of Bits to support open-source maintainers with AI-assisted vulnerability identification, validation, and fixing.
Patch the Planet initial participants: cURL, NATS Server, pyca/cryptography, Sigstore, aiohttp, the Go project, freenginx, Python, and python.org. The programme pairs AI-assisted security research with expert human review.
Tools: Codex Security, GPT-5.5-Cyber, Daybreak | Model: GPT-5.5-Cyber | Benchmark: 85.6% on CyberGym | Partners: Trail of Bits, HackerOne, Calif | Initial OSS projects: cURL, NATS Server, pyca/cryptography, Sigstore, aiohttp, Go, freenginx, Python, python.org
Best for: Security teams, open-source maintainers, and organisations interested in AI-assisted vulnerability detection, validation, and patching
Improved memory — ChatGPT Business, Enterprise, and Edu
When enabled, ChatGPT can now use relevant context from past chats to keep memory current and make responses more relevant as work changes. Users can review a memory summary, see sources below personalised responses, correct memory, delete referenced chats, mark sources as not relevant, turn memory off, or return to legacy Saved memories.
Enterprise begins with an approximately two-week early access period. Admins can turn on improved memory in Workspace settings; after early access, it turns on by default for eligible workspaces unless an admin opts out. This update does not affect Codex memory, and project-only memory remains contained within each project.
Plans: Business, Enterprise, Edu | Controls: memory summary, source visibility, source correction, referenced-chat deletion, not-relevant marking, memory off switch, legacy Saved memories | Enterprise rollout: two-week early access, admin opt-out available | Cost: no additional cost
Best for: Teams using ChatGPT across long-running workflows where context continuity matters but governance and user control remain important
Slack connector actions and large-paste handling
Enterprise and Edu workspaces can now enable Slack connector actions in ChatGPT — beyond searching Slack, ChatGPT can join a channel, create a reminder, upload a file, or update a user's Slack profile. Separately, the large-paste threshold was raised to 10,000 characters for all plans (up from 5,000 for Plus/Pro/Business), automatically converting oversized pastes into attachments.
Slack actions: admins review Slack under Apps and use Action control to manage which actions are available. Some actions require additional Slack OAuth scopes or Slack workspace/Enterprise Grid admin approval. Large paste: threshold now 10k characters across Free, Go, Plus, Pro, Business. Content can be moved back into the text field using "Show in text field."
Slack plans: Enterprise, Edu | Slack actions: join channel, create reminder, upload file, update user profile | Slack OAuth: some actions require additional scopes/admin approval | Paste threshold: 10k characters | Behaviour: automatic attachment conversion | Plans: Free, Go, Plus, Pro, Business
Best for: Enterprise and Edu teams using Slack as a core workflow layer; users pasting long reports, transcripts, or code into ChatGPT
Codex CLI 0.142.2
MCP tools now use tool search by default when supported, improving tool discovery. Adds macOS system proxy/PAC/WPAD support, dark-mode plugin logos, and richer safety-buffer UI. Fixes: Bedrock credential recovery, remote stdio MCP paths, model-visible remote image validation errors, PowerShell approval escalation, Code Mode metadata warnings.
Version: Codex CLI 0.142.2 | Install: npm install -g @openai/codex@0.142.2 | New: MCP tool search default, macOS proxy/PAC/WPAD, dark-mode logos, safety-buffer UI | Fixes: Bedrock credentials, remote stdio MCP paths, PowerShell escalation, Code Mode metadata warnings
Best for: Developers using Codex CLI, MCP tools, remote plugins, Amazon Bedrock integrations, or enterprise safety controls
Plans and Pricing
No major public ChatGPT or API pricing change for June 19–26. The most relevant enterprise change is the continued rollout of Global Admin Console controls: workspace, group, and user credit limits; billing and usage analytics; Codex and ChatGPT usage views; leaderboards; invoices; and overage limit settings.
Pricing: no new public model/API pricing change | Admin controls: workspace/group/user credit limits, billing analytics, usage export, Codex and ChatGPT usage views | Related rollout: Enterprise and Edu Global Admin Console updates
Best for: Enterprise and Edu admins managing AI spend, credit limits, and adoption analytics across ChatGPT and Codex
Gemini (Google)
Dateline: June 26, 2026 | Next update: July 3, 2026
The mandatory enforcement of an API security block on unrestricted keys is the dominant technical story this week. As of June 19, all Gemini API keys without explicit endpoint restrictions return 403 errors. Google also completed its scheduled phase-out of legacy visual preview engines, the legacy video generation stack entered its final countdown for a June 30 sunset, and Google executed its major Chat workspace redesign, launching the standalone "Ask Gemini" application interface.
⚠️ Unrestricted API key block — global enforcement
Unrestricted legacy keys are no longer accepted across the Gemini API ecosystem. This migration completely blocks open-access strings to mitigate unauthorised third-party scraping and prevent surprise cloud billing exploits. Developers using AI Studio or Google Cloud keys must manually restrict their active credentials exclusively to the Gemini API (generativelanguage.googleapis.com).
Keys lacking specific API configuration scopes will fail across all active integration endpoints. Image Model Sunset (June 25): Standard visual preview strings have been formally retired — production services must point to the GA version IDs (gemini-3.1-flash-image and gemini-3-pro-image). Veo Final Countdown: veo-2.0-generate-001, veo-3.0-generate-001, and veo-3.0-fast-generate-001 enter their final days before a hard termination on June 30 — migrate immediately to veo-3.1-generate-preview.
Models deprecated: gemini-3.1-flash-image-preview, gemini-3-pro-image-preview | Recommended replacement: gemini-3.1-flash-image, gemini-3-pro-image | Key dates: June 25 (image previews offline), June 30 (Veo 2.0/3.0 hard shutdown) | Error: unrestricted keys trigger 403 PERMISSION_DENIED | Fix: restrict keys to generativelanguage.googleapis.com in Google Cloud Console
Best for: All cloud developers and system architects — audit and configure API keys in the Google Cloud Console immediately to avoid application downtime
"Ask Gemini" — standalone Google Chat app
The conversational Workspace panel has transitioned to a permanent standalone app space named "Ask Gemini", completely replacing the traditional Google Chat sidebar tool. Standard prompt histories from the old sidebar do not automatically carry over to the new isolated application hub.
Standalone app interface is live, deprecating the older chat sidebar utility. New admin controls rolled out on June 16 allow IT administrators to programmatically enable or disable Temporary Chats and manage whether end-users have permission to clear local conversation logs. Data Retention Notice: sidebar prompt history does not migrate — manual user backup is required to preserve past conversations.
Platform: Google Chat Workspace | Deployment: mandatory server-side interface rollout | Data scope: non-migrating sidebar prompt strings | Admin variables: Temporary Chats visibility settings, user-driven delete permissions | Next phase: expanded contextual search across connected Drive repositories
Best for: Workspace IT administrators controlling user data privacy; enterprise teams moving complex prompt operations to a dedicated app view
Google Voice — AI note-taking for enterprise calls
"Take notes for me" feature is now operational for enterprise calls. The module automatically records audio, creates text transcripts, synthesises central conversation blocks, and sends structured action lists directly to the user's Gmail inbox at the conclusion of the call. Parallel Video Workflows: Google Vids users received updates on June 16 allowing concurrent processing of multiple timeline generations via Veo while strictly preserving consistent character tokens across continuous scenes.
Telephony engine: automated background transcription layer | Output delivery: structured JSON-to-text formatting sent via Gmail | Media engine: parallel rendering pipelines optimised for Veo multi-video timelines | Continuity parameters: strict character and environmental asset persistence
Best for: Business professionals requiring automated documentation for client calls; media teams iterating on long-form, multi-scene video concepts
DeepMind — AGI timeline and compute strategy
Google DeepMind CEO Demis Hassabis outlined a ~2030 AGI timeline ("plus or minus a year") at Cannes Lions and emphasised that Google's massive TPU infrastructure serves as a primary differentiator for drawing top-tier research talent to the Gemini ecosystem.
Projected AGI horizon: ~2030 | Core infrastructure: custom TPU cluster scaling | Research objective: enhancing multi-step reasoning in upcoming model architectures
Best for: Strategic enterprise partners and developers tracking long-term AI hardware scaling and foundational deep-learning roadmap expectations
Plans and Pricing
The promotional open access period for "Match Writing Style" and "Match Doc Format" in Google Docs has concluded — these features now require specialised Workspace AI add-on licences. API pricing tiers remain optimised for high-volume deployments following migration from legacy 2.0 to Gemini 3.5 Flash. Note the June 29 writerWithoutPrivateAccess Calendar API parameter going live.
Gemini 3.5 Flash: default engine, optimised for token throughput | Gemini 3.1 Flash-Lite: GA at $0.25/1M input tokens | Security restriction cost impact: eliminates phantom developer expenses from exposed keys | Next deadline: June 29 — writerWithoutPrivateAccess Calendar API parameter | June 30: Veo 2.0/3.0 hard shutdown
Best for: Budget administrators managing large-scale API operations; developers finalising system integrations before end-of-month deadlines
Microsoft Copilot
Dateline: June 26, 2026 | Next update: July 3, 2026
Three major developments defined this week: Copilot Cowork reached general availability; Anthropic's Claude was embedded inside Microsoft 365 Copilot across Word, Excel, PowerPoint, and Teams; and Microsoft introduced Scout, its first "Autopilot" persistent agent designed for continuous monitoring and proactive task execution.
Copilot Cowork GA, Claude integration, and Scout
Copilot Cowork — Microsoft's cloud-based agent that executes tasks asynchronously — is now available to all enterprise tenants. Admins must enable it explicitly and set usage-based billing caps. Claude is now embedded inside Microsoft 365 Copilot, offering structured reasoning alongside GPT-5.5 Instant across Word, Excel, PowerPoint, and Teams. Scout is Microsoft's first "Autopilot" agent: a persistent, proactively-executing agent positioned as the next evolution beyond Cowork.
Cowork: usage-based billing, off by default, enabled per tenant/group/user. Claude integration: available in Word, Excel, PowerPoint, and Teams; complements GPT-5.5 Instant with structured reasoning capabilities. Scout: persistent agent with continuous monitoring and proactive task execution; enterprise preview. Governance: Purview sensitivity labels extended to cover Cowork and Scout outputs.
Cowork: usage-based billing, off by default, tenant/group/user enablement | Claude: integrated into 365 apps, complements GPT-5.5 Instant | Scout: persistent agent, enterprise preview | Governance: Purview sensitivity labels extended to Cowork + Scout
Best for: Enterprises piloting agent workflows; teams needing asynchronous execution and structured reasoning in Microsoft 365
Copilot in Apps — Excel, Teams, Word, PowerPoint
Excel Plan Mode — preview changes before execution; Python integration expanded. Teams — call delegation, video recap summaries, meeting notes auto-fed into Copilot Notebooks. Word — Claude-powered writing suggestions; Copilot default editor mode. PowerPoint — FLUX.2 Flex image model; executive-summary slide generation; "Explain" for shapes and images.
Excel: Plan Mode, Python | Teams: recap + Notebooks | Word: Claude suggestions | PowerPoint: FLUX.2 Flex, Explain
Best for: Productivity teams needing structured previews, automated recaps, and design consistency
Governance and Compliance — Purview, Work IQ APIs, Federated Connectors
Federated Copilot Connectors GA — secure external app integration. Work IQ APIs GA — standardising agent intelligence across workflows. Purview expansion — sensitivity labels, DLP, and eDiscovery now cover Copilot and Loop content.
Connectors: federated, GA | Work IQ APIs: GA | Purview: extended coverage to Copilot + Loop
Best for: Enterprise IT and compliance teams managing Copilot adoption
Copilot Adoption Hub
Microsoft launched Copilot Adoption Hub, a guided rollout platform for enterprises. Includes campaign templates, training modules, and analytics dashboards to structure and track Copilot deployment.
Best for: Enterprise admins planning structured Copilot deployment across their organisation
Plans and Pricing
Cowork: usage-based billing, off by default. Claude integration included in Copilot Pro and Enterprise tiers. GPT-5.5 Instant available as lower-cost tier. Scout in enterprise preview with pricing TBD.
Cowork: usage-based billing, off by default | Claude: included in Copilot Pro + Enterprise | GPT-5.5 Instant: lower-cost tier | Scout: enterprise preview, pricing TBD | Models in Copilot: Claude + GPT-5.5 Instant | Policy: unified Copilot Chat governance
Best for: Enterprises setting billing caps; developers leveraging Claude + GPT-5.5 Instant mix in Copilot APIs