KI-Modelle: Release Notes

• Added onFailure: "block" for command and HTTP hooks: a hook that can't start, times out, or exits with an unexpected code blocks the action instead of letting it through
• Added Program Status Protocol (OSC 7501) support: terminals that implement it can show whether Claude Code is working, waiting on you, or done
• Added quoted text to the /copy picker, so a drafted message copies without its > markers
• Added a warning to claude plugin install, enable, disable and marketplace add when the settings file they write to does not load
• Added a line on stderr, when it is a terminal, that says what a claude -p run is waiting for when it stays open after its last turn
• Added support for timeouts.upstream_ttfb_ms on the Claude apps gateway's Bedrock, Vertex, Foundry and other cloud upstreams: a value you set now limits how long a stream may take to start there, after which it fails over or gets a 502
• Added a "Backgrounding cancelled" message when you stop the turn while ← is waiting for the current tool to finish
• Added an optional models list to every Claude apps gateway upstream: only the listed models are sent there, on failover too, and one * in an entry is a wildcard
• Added support for forceLoginMethod: "gateway" and forceLoginGatewayUrl in your own user settings on machines with no managed settings, so /login opens on that Claude apps gateway …

• Added Claude Haiku 5.5 (claude-haiku-5-5), now the default Haiku model on the Anthropic API — 1M context, $0.10/$0.50 per Mtok ($0.50/$2.50 for prompts over 100K)
• Added agentType to the subagentStatusLine payload, so scripts can tell custom subagent types apart
• Added isDeferred to $.tool.register for mods: false lists the tool's schema in the prompt from the start instead of behind tool search
• Fixed Claude sometimes treating its own last actions before a context compaction as done after it, and retracting or redoing finished work
• Fixed a memory leak where an HTTP MCP connection kept every request it had sent until it closed
• Fixed a message sent while Claude was working being lost when ← moved the session to the background; if a queued message can't move, ← now stays put and says so
• Fixed /model effort ←/→ wrapping past the highest or lowest level, which could accidentally save Low as a model's default effort
• Fixed /tui disconnecting Claude in Chrome in a session started with --chrome, and ignoring --no-chrome
• Fixed Claude being told to continue or message subagents with SendMessage in sessions, including resumed ones, where a host, a permission rule or a --tools list removed that tool
• Fixed subagents and --agent sessions being told a built-in tool was disabled for the whole session when only their own tool list left it out …

• We've lowered the price of prompt cache reads on Claude Sonnet 5.5 from $0.20 USD to $0.10 USD per million tokens: 0.05x the base input price instead of 0.1x. Cache writes and all other prices are unchanged. See Prompt caching pricing.
• We've launched Claude Haiku 5.5 (claude-haiku-5-5), our most capable model tuned for high-volume and latency-sensitive work. It has a 1M token context window, 128k max output tokens, and adaptive thinking with the effort parameter. It's available on the Claude API, Claude in Amazon Bedrock, Claude Platform on AWS, Claude on Google Cloud, and Claude in Microsoft Foundry. See What's new in Claude Haiku 5.5. …

SDK Releases

• The Firebase Apple SDK (v13.0.0) is now available. For more details, see the Firebase Apple SDK release notes. To install this SDK, see Add Firebase to your app.

• The Firebase Android SDKs for A/B Testing, AI Logic, Analytics, App Check, App Distribution, Authentication, Cloud Firestore, Cloud Functions, Cloud Messaging, Cloud Storage, Crashlytics, Crashlytics NDK, Dynamic feature modules support, In-App Messaging, Installations, Performance Monitoring, Firebase PNV, Realtime Database, Remote Config, and SQL Connect have been updated. For more details, check out the latest Android Release Notes. To get started with Firebase in Android, see Add Firebase to your Android Project.

• The Firebase Android BoM (Bill of Materials) has been updated to v35.0.0. Learn how to add SDKs to your app using the Firebase BoM.

• The Firebase JavaScript SDK (v13.0.0) is now available. For more details, see the Firebase JavaScript SDK release notes. To install the SDK, see Add Firebase to your JavaScript Project.

New Features

• GPT-6.1 Sol is now the default model in the bundled and Amazon Bedrock catalogs. (#49318, #49339)
• Amazon Bedrock supports multi-agent V2 and Ultra reasoning on compatible models; Bedrock Mantle also accepts AWS GovCloud regions. (#49345, #49813)
• Sign in to MCP servers from an active terminal session with /mcp login <name>. (#49290)
• Choose your microphone, speaker, and microphone input channels for voice conversations, with preferences saved locally. (#49437, #49836)
• Daybreak is opt-in through --enable cli_daybreak or features.cli_daybreak=true; daybreak=true alone is insufficient. By default, controls and indicators are hidden, /daybreak is unavailable, and automatic Cyber routing is omitted—even for saved Daybreak threads. Saved preferences remain intact. Opt-in routing requires eligible ChatGPT sign-in, the OpenAI provider, and advertised model/program support. (#49856, #49858, #49859, #49861, #51207)
• Select a Cyber access program per turn with codex exec --cyber-access-program or the TypeScript SDK’s cyberAccessProgram option. The explicit exec override remains available with cli_daybreak disabled and leaves the saved choice unchanged. (#49939, #51207)

Bug Fixes

• Approved filesystem escalation can now grant broader write access while preserving denied reads and network restrictions. Background tasks retain their originating turn’s permissions. (#49353, #49880) …

August

Imagine image API updates

• Auto quality. The quality parameter on grok-imagine-image-2.0 now accepts auto, and the default when quality is omitted has moved from medium to auto. Auto currently uses low for image generation and medium for image editing. Images are billed at the quality they are served at. Pass low or medium explicitly to pin a specific quality. See Image Generation.
• Five reference images. Image editing now accepts up to 5 source images per request (was 3). See Multi-Image Editing.
• New aspect ratios. Image generation and editing accept 21:9 (cinematic widescreen) and 5:2 (wide banners). See Image Generation.

Developer and creator Tom Shaw sat down with Meta’s Head of Infrastructure, Santosh Janardhan, to talk about Meta’s approach to building infrastructure to power the future of AI.

Santosh explained why Meta is more than just a software company, how AI is different from any other technology, and why data centers are essential for building AI. 

They’ll cover questions like: 

• How does Meta power its AI infrastructure? 
• What is a gigawatt of energy? 
• How does Meta decide what kind of chips to use for their data centers? 
• Why is Meta so focused on AI right now? 
• What are the benefits of building your own data centers?

and more. 

The post Why Data Centers Are Such a Big Part of Meta’s AI Approach appeared first on Meta Newsroom.

Keeping young people safe on our apps is among the most important work we do, and it never stops.

Sophisticated bad actors who try to exploit children online don’t stay in one place. These determined predators move across different platforms, and change their tactics to evade detection. Staying ahead requires constant adaptation and that is why we are continuously evolving our tools, sharing signals across the industry, and reporting offenders to law enforcement so they can be held accountable.

How We Responded

Whenever we detect or are made aware of potential child sexual abuse or exploitative content on our platform, we investigate it. If the content violates our policies, we remove it and in cases where the signals indicate malicious sharing by predators, we disable the accounts behind it. We then report the content to the National Center for Missing & Exploited Children (NCMEC), in line with applicable law. But we haven’t stopped there. We have been doing strategic network disruptions on child safety for years – our team of experts including former FBI investigators who conduct manual investigations to proactively uncover, remove and report networks of predatory accounts. We continue to investigate how these networks behave so we can act against them at a network level, not just account by account. …

Feature

Anthropic's Claude Haiku 5.5

Claude Haiku 5.5 is available in Model Garden.

Fixed

CodeMender updates (v0.13.0)

This release introduces updates to CodeMender:

• Consistent finding deduplication: Improved finding deduplication consistency across repeated cm find, --deep, --diff, and --parallel scans.
• Bug fixes:
  • Fixed an issue where deduplicated findings could appear in cm report --status DISMISSED.
  • Improved local state reliability during concurrent cm find --diff and cm find --deep scans.
  • Updated streaming error handling so non-retryable errors fail immediately instead of retrying.
  • Ensured local session state and logs are saved cleanly when a session is interrupted (Ctrl+C).

For more information, see the CodeMender documentation.

Feature

Gemini Enterprise: Pay-as-you-go usage above quota enabled for Frontline, EDU, and Emerging Market editions

You can enable pay-as-you-go usage above quota for additional Gemini Enterprise license-based editions:

• Frontline editions: Frontline Starter and Frontline Worker
• EDU editions: EDU and EDU Pro
• Emerging Market editions: Standard Emerging Market, EDU Emerging, EDU Pro Emerging, and EDU Gov Emerging

When you enable Pay-as-you-go usage on an invoiced Cloud Billing account with an active subscription, users can continue using features at pay-as-you-go rates after reaching their pooled quota limits.

For more information, see the following resources:

• Quotas and overages
• Configure overages
• Overview of overages and spend controls