Dateline: April 09, 2026 | Next update: April 16, 2026
Over the past week, Anthropic has focused on enterprise agent infrastructure, developer tooling polish, and a landmark model preview. Claude Code shipped multiple point releases; Claude Managed Agents launched; and Claude Mythos became available as an invitation-only research preview.
Claude / Anthropic
Claude Mythos Preview — frontier research model
Anthropic's most powerful model to date is now in the hands of select early-access partners. Internally described as a step-change in capability, Mythos is being trialled exclusively for defensive cybersecurity work before any broader release.
Claude Mythos Preview is live as a gated research preview under Project Glasswing. Access is by invitation only and restricted to defensive cybersecurity use cases. No public release date has been announced.
Access: invite-only | Use case restriction: defensive cybersecurity only | Anthropic flagged an unprecedented cybersecurity risk profile requiring controlled rollout
Best for: Cybersecurity researchers and red teams — invitation required
Claude Managed Agents — hosted agent infrastructure
Anthropic launched a fully managed hosting layer for Claude-powered agents, removing the infrastructure burden from enterprise teams. Managed Agents handles scaling, monitoring, and deployment so organizations can focus on agent logic rather than servers.
Claude Managed Agents launched — managed hosting, automatic scaling, and built-in monitoring for Claude agents. No infrastructure setup required. The /claude-api skill in Claude Code has been updated to cover Managed Agents alongside the direct API.
Complements existing API | Targets enterprises running multi-agent workflows at scale | Anthropic manages compute and uptime SLAs
Best for: Enterprise teams deploying autonomous agents in production
Claude Messages API on Amazon Bedrock — research preview
Anthropic's first-party Messages API shape is now available directly on Amazon Bedrock infrastructure, with zero operator access. AWS customers can call Claude with the same request format they already use, without routing through Anthropic's servers.
Messages API available on Amazon Bedrock as a research preview at /anthropic/v1/messages. Runs on AWS-managed infrastructure with zero operator access. Currently available in us-east-1.
Endpoint: /anthropic/v1/messages on Bedrock | Region: us-east-1 | Same request shape as first-party Claude API | Zero operator access | Contact AE for access
Best for: AWS-native teams, regulated industries needing data residency on AWS
Claude Code — agentic coding tool
Claude Code shipped several point releases this week, adding new learning tools, performance improvements, and bug fixes. The default effort level has also been raised for most users.
Added /powerup — interactive in-terminal lessons teaching Claude Code features with animated demos. Default effort level raised from medium to high for API-key, Bedrock/Vertex/Foundry, Team, and Enterprise users. Amazon Bedrock support via Mantle added (CLAUDE_CODE_USE_MANTLE=1). Write tool diff speed improved 60% for large files. MCP tool result size cap raised to 500K via annotation override.
New env var: CLAUDE_CODE_USE_MANTLE=1 for Bedrock via Mantle | /effort command controls level (low/medium/high, now displayed as ○ ◐ ●) | MCP: _meta["anthropic/maxResultSizeChars"] annotation allows up to 500K result size | disableSkillShellExecution setting added | Plugins can now ship executables under bin/
Best for: Developers, CI pipelines, terminal-based agentic coding
Claude Developer Platform — API updates
Key migration deadlines are approaching. The 1M context window beta for older Sonnet models retires April 30. Claude Haiku 3 retires April 19. Both require migration to newer models.
Haiku 3 (claude-3-haiku-20240307) retires April 19 — migrate to Haiku 4.5 now. 1M token context beta for Sonnet 4.5 and Sonnet 4 retires April 30 — migrate to Sonnet 4.6 or Opus 4.6 which include 1M context natively. Models API now returns capability fields: max_input_tokens, max_tokens, and a capabilities object.
Haiku 3 retirement: April 19, 2026 | Context-1m beta retirement for Sonnet 4/4.5: April 30, 2026 | 300k Batches API output: use header output-300k-2026-03-24 | Models API: GET /v1/models now returns capabilities object
Best for: API developers — action required if using deprecated models or beta headers
Core models — no new releases
No new model versions shipped this week. Sonnet 4.6 remains the default balanced model; Opus 4.6 remains the flagship. Haiku 4.5 is the recommended fast/low-cost option following Haiku 3's deprecation.
No model releases this week. Stability week for the current generation.
Sonnet 4.6: $3/$15 per M tokens | Opus 4.6: $5/$25 per M tokens | Both support 1M token context natively
Best for: Stable usage — no action needed unless on deprecated models
Plans and Pricing
No changes to consumer or API pricing this week. Managed Agents pricing: contact enterprise sales.
ChatGPT / OpenAI
Dateline: April 09, 2026 | Next update: April 16, 2026
ChatGPT is an AI system developed by OpenAI that helps with writing, research, and task automation.
GPT-5.3 Standard — the default model
Improved reliability and tool usage.
Reduced hallucination. (April 5, 2026)
Context ~128k | Output ~4k–8k
Best for: General use
GPT-5.3 Pro — high-reasoning model
Improved reasoning consistency.
Better multi-step reasoning. (April 6, 2026)
Context ~200k (est.)
Best for: Deep analysis
GPT-5.3 Mini — fallback model
Fast responses and fallback routing.
More aggressive fallback. (April 4, 2026)
Context ~64k
Best for: Speed
Agent Mode
Handles multi-step workflows.
Better ambiguity handling. (April 7, 2026)
Cloud-based execution
Best for: Task automation
Deep Research
Combines browsing and reasoning.
Improved sources. (April 6, 2026)
High latency
Best for: Research
Memory & Projects
Persistent memory across chats.
Better relevance. (April 5, 2026)
Global + project memory
Best for: Personalization
Advanced Voice Mode
Improved conversational voice.
Better turn-taking. (April 4, 2026)
Mobile-first
Best for: Voice interaction
Plans and Pricing
No changes this week. Plans unchanged.
Gemini
Date: April 9, 2026 | Next update: April 16, 2026
This week's focus has been on expanding Gemini Enterprise capabilities through new data connectors and improving collaborative features within Workspace. For developers, the Gemma 4 series has launched, providing new open models for flexible implementation.
Core models — Gemini 3.1 series
Gemini 3.1 Pro: Following the launch of "High Effort" mode on April 2, the model continues to serve as the flagship for complex logic and debugging.
Gemini 3.1 Flash Lite: Now in General Availability (GA), this model provides 99.9% uptime for high-volume enterprise tasks.
Gemma 4 (new): Released on April 2, 2026, the gemma-4-26b and gemma-4-31b instruction-tuned models are now available via the Gemini API and AI Studio.
Inference tiers: New Flex and Priority tiers were introduced on April 1, allowing developers to optimize for either cost or latency.
Google Slides integration
Gemini received a major upgrade on April 3, 2026, enabling the generation of fully editable, brand-matched slides that analyze existing decks for visual consistency.
Best for: Creative presentation work
Google Vids — AI avatars
Users can now create and refine custom AI Avatars using Nano Banana 2 (Gemini 3.1 Flash Image). A new feature allows directing these avatars to interact with specific objects while maintaining consistent faces and voices across scenes.
Best for: Enhanced Google Slides and Vids AI Avatars
NotebookLM Enterprise
As of April 2, 2026, autocomplete is available for email addresses and group names when sharing notebooks.
Google Meet — speech translation
Speech translation is now rolling out to Android and iOS mobile devices as of April 8, 2026, providing near-real-time audio translation.
Best for: Global collaboration
Gemini Enterprise connectors
Jira and Confluence federated connectors reached General Availability on April 3, 2026. Google Drive now supports analyzing CSV, PDF, PPTX, and XLSX files directly within the connector. New actions in preview for Gmail, Drive, GitHub, HubSpot, and Monday added March 31, 2026.
Plans and migration deadlines
Whisk to Flow migration: This is the final month for the transition. All users must migrate assets from Whisk to Flow before the April 30, 2026 deadline, or they will be permanently deleted.
New add-on: Google introduced the AI Expanded Access add-on for Workspace Business users in the U.S. and Canada, allowing direct user-initiated purchases of advanced AI features.
Service status: Intermittent issues with Gemini API Batch processing were detected on April 3 and April 6, but both incidents have been resolved.
Microsoft Copilot
April 2–9, 2026
Copilot for Education — six AI-driven teacher tools
Lesson planning and classroom activities now benefit from AI support, including alignment with standards in 35 countries, fill-in-the-blank exercises, and matching activities with leaderboards.
Standards alignment engine (35 countries) | "Add Supporting Examples" module | Fill-in-the-blank generator with analytics | Matching activity builder with gamification
Copilot Researcher — multi-agent reliability checks
Reports now draw from hundreds of sources with improved accuracy.
Multi-agent validation pipeline | Enhanced source reliability scoring
Copilot Cowork — shared AI-driven drafting
Teams can co-create research and documents using Copilot with Anthropic integration.
Multi-user drafting environment | Anthropic collaboration layer
Copilot Chat — video recap feature
Meeting summaries now include narrated highlight reels with relevant clips.
Clipchamp integration | Audio recap expanded to more languages
Copilot Notebooks — expanded export formats
Reports can now be exported as PowerPoint, PDF, infographic, or audio.
Multi-format export engine | Enhanced layout options
Dragon Copilot — reduced per-user license pricing
Dragon Copilot is now more affordable for partners worldwide.
Reduced per-user license pricing | Applies globally
Copilot Chat — licensing changes
Copilot Chat remains in Outlook but will be removed from Word, Excel, PowerPoint, and OneNote unless users have a paid Copilot license.
Outlook grounding (inbox + calendar) | Restricted access in other apps
Plans and Pricing
No other changes this week. Plans unchanged | API token billing.