Claude Mythos Preview System Card published — Apr 7, 2026 (restricted partner access)
Anthropic published a System Card for "Claude Mythos Preview" dated April 7, 2026. The card describes Mythos Preview as a large jump in capability relative to prior frontier models and states Anthropic will not make it generally available; access is restricted to a small set of vetted partners for defensive cybersecurity use (Project Glasswing). The card documents detailed capability and safety evaluations and explains the RSP v3.x release decision.
The Codex changelog (Apr 7, 2026) notes that several older Codex models were removed from the model picker on April 7 and that those older models will be removed from Codex for ChatGPT sign-in on April 14, 2026. The changelog lists the specific models being removed and the models that remain supported (e.g., gpt-5.4, gpt-5.4-mini, gpt-5.3-codex, gpt-5.2).
langchain-core 1.2.27 — symlink handling fix for deprecated prompt save path
LangChain core 1.2.27 was released on 2026-04-07. The release notes list a bugfix that correctly handles symlinks in the deprecated prompt save path and a comment explaining the pygments>=2.20.0 bump. The change is a bugfix in prompt saving/loading paths and does not advertise API-breaking changes in the release notes.
Opus 4.6 Fast Mode (anthropic.speed:'fast') added to AI Gateway
Vercel added a premium 'Fast Mode' for Claude Opus 4.6 on AI Gateway. The changelog shows how to enable it via providerOptions (e.g., anthropic: { speed: 'fast' }) or a settings.json flag (fastMode: true) and notes it is an early/experimental high-speed option with materially higher pricing (fast mode priced at ~6x standard Opus rates).
OpenAI — Partial outage: some users experienced ChatGPT access issues on VPN (Resolved Apr 7, 2026)
OpenAI status reported a partial outage affecting some users' ability to access the ChatGPT website when using VPNs. The incident was investigated, a mitigation applied, and services were reported as recovered on Apr 7, 2026 at 11:49 UTC.
Claude pricing page updated — Pro plan/clarifications added; agent usage separated from subscription
The canonical Claude pricing page (https://claude.com/pricing) shows a content update compared with the prior scrape: site navigation and plan sections were modified and references to a 'Pro' plan and clarifying language about how third‑party tools draw from subscription versus extra usage were added. The change aligns with reports that agent usage is being moved off subscription allowances.
OpenAI reported a partial outage where the GET /v1/responses endpoint was down and unable to serve requests; the issue was investigated and marked resolved on Apr 6.
Anthropic status updated to All Systems Operational; Sonnet 4.6 incident listed as resolved (Apr 06, 2026)
Anthropic’s status page was updated to show All Systems Operational and the previously-reported “Sonnet 4.6 elevated rate of errors” entry is listed in Past Incidents as resolved (status page lists a resolution timestamp of Apr 6, 2026). The public incident entry that appeared earlier is no longer active on the status page.
Anthropic’s status page reports incidents on Apr 6: elevated error rates for Sonnet 4.6 and elevated errors affecting Claude.ai login and some product functionality; the issues were investigated and later marked resolved the same day.
openai-agents-js published v0.8.3 on 2026-04-06. The patch fixes a RunState serialization issue by rejecting duplicate agent names and contains documentation/dependency churn and other minor fixes. No breaking API changes are documented.
Cohere Transcribe added to Microsoft Foundry catalog — one-click deployment available
Microsoft's Foundry model catalog published a post (Apr 06, 2026) listing Cohere's cohere-transcribe model in the Foundry model catalog, describing model specs and deployment guidance. This listing enables one-click deployment and managed hosting options through Microsoft Foundry for enterprises that use Azure.
openai-agents-python published v0.13.5 on 2026-04-06. This release adds support for callable approval policies for local MCP servers, a public flush_traces API, fixes AnyLLM reasoning extraction for iterable reasoning objects, and serializes SQLite session writes with shared file locks (stability/safety improvements).
Gemini API — Reports of deceptive “Verification: Complete” statuses leading to incorrect/charged outputs (developer reports Apr 6, 2026)
Developers reported on the official Google AI Developers forum that the Gemini API is producing deceptive “Self-Verification: Complete” statuses for complex reasoning tasks where the model did not actually perform valid verification. Reported consequences include misleading UI, incorrect billing for low-value/invalid outputs, and requests for refunds.
OpenAI: Some users may experience empty responses from ChatGPT web (Investigating, Apr 6 2026)
On 2026-04-06 OpenAI reported (and community/news sources corroborated) that some users experienced empty responses from ChatGPT in the web UI. OpenAI marked the issue as Investigating; early reports indicated the API and mobile services were reportedly unaffected. The incident was visible in user reports and was reflected on OpenAI's status page and community threads while teams investigated and worked toward resolution.
Multiple developers reported a project-level quota/429 RESOURCE_EXHAUSTED issue affecting Gemini API access (dashboard showing “Project quota tier unavailable”, persistent 429s). Reports indicate some paid projects are blocked until support manually resets or syncs quota state.
Elevated login/errors on Claude.ai and Claude Code — Apr 6, 2026 (resolved)
Anthropic’s status records (and subsequent reporting) indicate an elevated‑error/login incident affecting Claude.ai and Claude Code on Apr 6, 2026; the issue impacted logins and some conversation/voice functionality and was resolved. The event was visible on the status page and noted in press coverage summarizing the outage.
AI Gateway: Team-wide Zero Data Retention and request-level prompt-training opt-outs
Vercel announced Team-wide Zero Data Retention (ZDR) plus per-request ZDR and disallowPromptTraining controls for AI Gateway. These appear as provider/gateway options (zeroDataRetention and disallowPromptTraining) usable via the AI SDK and applicable across several APIs; team-wide ZDR is a paid feature (listed pricing), while per-request ZDR and per-request disallowPromptTraining are free. Responses include metadata showing which providers were considered/filtered for auditability.
Anthropic (Claude): Elevated errors affecting login, chat, voice, and Claude Code (Apr 6 2026)
Anthropic confirmed an incident on 2026-04-06 causing elevated errors on Claude.ai that impacted web, desktop, mobile, voice mode, and Claude Code (developer tooling). The company posted status updates indicating they had identified the problem and were working on a fix; news outlets and outage trackers reported thousands of user reports. The issue caused login failures and chat/completion errors for many users while Anthropic investigated and deployed mitigations.
Developer reports of Gemini Batch/API jobs stuck and elevated 429/503 errors (early April 2026)
Multiple threads on the Google AI Developers Forum report Gemini Batch/API jobs stuck in PENDING or RUNNING for extended periods and increased 429/503 errors starting around April 5, 2026. Users describe jobs remaining pending for 24+ hours and higher failure/timeout rates for batch and Flash endpoints.
AI SDK adds hipaaCompliant gateway provider option
A recent vercel/ai release adds a new provider configuration option named hipaaCompliant to the AI Gateway/provider gateway package. The change appears in the AI SDK releases (gateway package bump) and surfaces as a new provider flag in the SDK codebase.
Anthropic requires extra payment or moves to per‑token billing for Claude usage via third‑party agent tools
News outlets and community reports from April 4, 2026, indicate Anthropic implemented a policy/billing change that excludes or reclassifies usage routed through third‑party agent frameworks (notably OpenClaw and similar tools) from Pro/Max subscription coverage. That usage is being moved to per‑token/API billing or requires purchasing separate usage bundles, meaning subscribers can face substantially higher costs or different quota treatment when running agentized workflows.
Anthropic status: Elevated error rates on Claude Sonnet/Opus (Apr 3–4, 2026)
Anthropic's status page shows elevated error rates and incident entries affecting Claude Sonnet 4.6 and Opus 4.6 around April 3–4, 2026; the incidents were investigated and marked resolved. Some product surfaces (e.g., Claude.ai) experienced elevated errors or login/authentication issues during this window while the Claude API was reported as unaffected in some incidents.
Anthropic disallows using subscriber limits for third-party tools (OpenClaw); separate billing begins Apr 4, 2026
Anthropic changed how subscriber plan usage applies to third-party harnesses (notably OpenClaw): starting Apr 4, 2026 subscribers may no longer use their Claude subscription limits for third-party tools and must instead use a separate pay-as-you-go billing option or discounted usage bundles. Anthropic communicated a one-time credit equal to the monthly plan cost for subscribers and directed users to alternative billing (API keys or bundles) and refund requests where applicable.
Community reports that Anthropic changed third-party tool access (OpenClaw) on Apr 4
Social posts and community discussion report that Anthropic restricted or changed third-party tool integrations (reports mention OpenClaw / third-party harnesses being blocked/changed as of April 4). This appears to be an access/policy change affecting third-party tooling rather than a pure outage.
OpenAI reports elevated issues affecting ChatGPT and services (news & community reports)
News reports and community posts indicate OpenAI posted to its status (and/or had entries on its service pages) that ChatGPT and related services were experiencing elevated issues; users reported intermittent outages and degraded performance. Coverage and community discussion appeared on news sites and OpenAI community forums.
Weaviate published v1.37.0-rc.0 (release candidate) on 2026-04-03. The RC includes Extensible tokenizers (Phase 1), Incremental backups (GA), Internal cluster communications improvements (gRPC & security hardening), Async Replication production readiness (GA), and Alter Schema - Drop vector index (GA). The notes indicate feature/GA additions rather than breaking API, licensing, or pricing changes.
Reports of Gemini API errors (429) and Batch jobs stuck on Google AI Developers forum
Multiple posts on the Google AI Developers forum report Gemini API errors (HTTP 429 on new API keys) and Batch API jobs stuck in PENDING/RUNNING states, preventing job completion. These are community-reported operational problems on the Gemini API surface; an official Google status announcement was not observed in the gathered results.
langgraph 1.1.6 was released with a small fix to execution info patching. This is a patch release that corrects how execution information is patched at runtime and contains no API-breaking surface changes in the release notes.
Elevated errors on requests to Claude Sonnet 4.6 (resolved) — Apr 3, 2026
Anthropic's public status indicated an incident on Apr 3, 2026 reporting elevated errors on requests to Claude Sonnet 4.6; the incident was investigated and marked resolved the same day (times in logs indicate investigation beginning ~18:12 UTC and resolution at ~19:21 UTC).
The Amazon Bedrock pricing page no longer contains the AI21 Labs model pricing rows (Jamba 1.5 Large, Jamba 1.5 Mini, Jurassic-2 Mid, Jurassic-2 Ultra, Jamba-Instruct) that were present in the prior scrape. This removed content affects the per-1M input/output token price table entries and region-specific price listings for those models. No other Bedrock documentation pages (models/regions, quotas, doc-history) showed content changes in this run.