Post-I/O week wind-down: xAI’s Grok Build CLI enters beta, OpenAI ships Codex CLI 0.134.0, Gemini Spark opens to broader US subscribers, and two AI infrastructure companies report active funding rounds.

Run: 24–27 May 2026 (72h) · 24 items reviewed → 7 published · 2 verified · 5 secondary · 0 rumor · 35% exploration · Run timestamp: 2026-05-27

TL;DR

• OpenAI Codex CLI 0.134.0 — adds local conversation search and per-server MCP environment targeting with OAuth for streamable HTTP servers. (→ OpenAI Codex CLI 0.134.0 adds conversation search and MCP OAuth)
• xAI Grok Build CLI — xAI’s coding agent CLI entered early beta on May 25 for SuperGrok and X Premium Plus subscribers; runs on Grok 4.3 beta with a 2M-token context window. (→ xAI Grok Build CLI enters early beta for SuperGrok subscribers)
• Gemini Spark broader US access — Google’s 24/7 cloud AI agent expands beyond trusted testers to all Google AI Ultra subscribers ($100/month) starting this week. (→ Gemini Spark expands to broader US access at $100/month)
• Suno AI raising $250M+ — Bond Capital leading a round that would value the AI music startup at $5B, nearly double its November 2025 valuation. (→ Suno AI raising $250M+ at $5B valuation)

Items

OpenAI Codex CLI 0.134.0 adds conversation search and MCP OAuth

Source: https://github.com/openai/codex/releases · OpenAI / GitHub · 2026-05-26 Verification: T2 verified · changelog · dev-tools

OpenAI tagged Codex CLI 0.134.0 on May 26, 2026. The release adds the ability to search across local conversation history with case-insensitive content matching and result previews. MCP setup gains per-server environment targeting and OAuth options for streamable HTTP servers, and read-only MCP tools can now run concurrently when they advertise readOnlyHint. The profile system was updated to make --profile the primary selector across CLI, TUI permissions, and sandbox flows, with legacy configurations now rejected through migration guidance. Extension and hook context was expanded to include conversation history for extension tools and subagent identity in hook inputs.

Why it matters for automation/productivity: The MCP OAuth addition means Codex agents can now authenticate to streamable HTTP MCP servers without workarounds, enabling richer integrations with SaaS tooling. Concurrent read-only tools reduce round-trip latency for agent pipelines that call multiple read-heavy MCP servers in parallel.

Key claims:

• Release date: 2026-05-26 → https://github.com/openai/codex/releases (T2)
• Per-server MCP environment targeting and OAuth for streamable HTTP → GitHub release notes (T2)
• Concurrent read-only MCP tools when readOnlyHint is set → GitHub release notes (T2)

Cross-references:

• https://developers.openai.com/codex/changelog (T2, full Codex changelog)
• https://releasebot.io/updates/openai/codex (T3, corroborating)

xAI Grok Build CLI enters early beta for SuperGrok subscribers

Source: https://cryptobriefing.com/xai-grok-build-coding-agent-beta/ · CryptoBriefing · 2026-05-25 Verification: T2 secondary · announcement · dev-tools Tier nuance: x.ai primary returned HTTP 403 in this run. Confirmed via CryptoBriefing (T2), Engadget, and The Tech Outlook, all with consistent launch date and feature details. Primary URL: https://x.ai/news/grok-build-cli

xAI launched Grok Build CLI in early beta on May 25, 2026, available exclusively to SuperGrok and X Premium Plus subscribers. The CLI operates from the terminal — users open it inside a repository, describe a task, review the plan, approve the diff, and let the agent execute. Key features include Plan Mode (requires user approval before changes are applied), Arena Mode (runs multiple models in parallel for side-by-side comparison), parallel subagents, worktree support, ACP support, and the ability to read Claude Code and AGENTS.md instruction files. The underlying model is Grok Build 0.1, which xAI previously made available via API on May 20 at $1/$2 per 1M input/output tokens with a 256K context; the CLI version exposes a 2M-token context window.

Why it matters for automation/productivity: Grok Build is xAI’s direct entry into the agentic coding tool market dominated by Anthropic Claude Code and OpenAI Codex CLI. For teams already on the xAI/SuperGrok stack, the CLI provides an immediately testable coding agent alternative. Arena Mode differentiates from competitors by enabling parallel model comparisons within a single task.

Key claims:

• Early beta launch: 2026-05-25 → CryptoBriefing, The Tech Outlook, Engadget (T2-T3, consistent)
• Access: SuperGrok and X Premium Plus subscribers, $300/month → CryptoBriefing (T2)
• Context window: 2M tokens (CLI version) → CryptoBriefing (T2)
• API version: grok-build-0.1, $1/$2 per 1M tokens, 256K context, available May 20 → Codersera benchmark roundup (T3)

Cross-references:

• https://www.engadget.com/2173482/xai-coding-agent-grok-build/ (T2, corroborating)
• https://www.thetechoutlook.com/new-release/software-apps/xai-launches-an-early-beta-of-grok-build-available-for-all-supergrok-and-x-premium-plus-subscribers/ (T3, corroborating)
• https://codersera.com/blog/grok-build-vs-claude-code-vs-codex-cli-2026/ (T3, comparative benchmarks — vendor SWE-bench figures; no independent audit as of this run)

Caveats: xAI primary source returned 403; verified through multiple independent outlets. Grok Build 0.1 vendor SWE-bench score of 70.8% belongs to the deprecated grok-code-fast-1; no published benchmark for grok-build-0.1 as of May 27. xAI reported internal staffing turbulence following the SpaceX merger; senior departures noted in press but not confirmed by xAI officially. Pricing at $300/month is high relative to Claude Code ($20 Claude Pro) and Codex CLI.

Gemini Spark expands to broader US access at $100/month

Source: https://blog.google/innovation-and-ai/products/gemini-app/next-evolution-gemini-app/ · Google · 2026-05-19 (broader rollout week of 2026-05-26) Verification: T3 secondary · deployment milestone · workflow-automation Tier nuance: The rollout milestone is based on multiple corroborating sources citing the week of May 26 expansion. No Google blog post announces the broader access date explicitly; it was signaled in the I/O announcement and confirmed by Tech Times (May 25) and findskill.ai rollout guide.

Google’s Gemini Spark — a 24/7 cloud AI agent that executes tasks across Gmail, Calendar, and third-party apps while running in the background — began rolling out to all Google AI Ultra subscribers in the US during the week of May 26, 2026, expanding beyond the trusted-tester cohort that received access after Google I/O on May 19. Google reduced the AI Ultra subscription price from $250 to $100 per month, making Spark the flagship feature of the new pricing tier. Spark can complete multi-step tasks without user intervention for each step: scheduling, email drafting, cross-app lookups, and document actions are described in the launch materials. Google’s own in-product warning states the agent may complete sensitive actions — including sharing information or initiating purchases — without per-action confirmation, depending on task type.

Why it matters for automation/productivity: Gemini Spark is the first major consumer 24/7 cloud AI agent from a hyperscaler to reach general subscriber access at under $150/month. For teams evaluating autonomous workflow tools, this creates a direct comparison point with Anthropic Claude Cowork and OpenAI Operator. US-only at this rollout; EU launch is pending AI Act compliance review (analysts estimate Q3 2026 at earliest).

Key claims:

• Subscription price: $100/month (reduced from $250/month) → Google I/O announcement via CNBC (T2)
• Broader US access starting week of May 26 → Tech Times (2026-05-25), findskill.ai rollout guide (T3)
• Available to all Google AI Ultra subscribers in the US → multiple corroborating sources (T3)

Cross-references:

• https://www.techtimes.com/articles/317144/20260525/gemini-spark-googles-24-7-cloud-ai-agent-now-executes-tasks-third-party-apps.htm (T3, corroborating, May 25)
• https://www.cnbc.com/2026/05/19/google-ai-ultra-gemini-spark-omni.html (T2, I/O announcement with pricing)
• https://theaiinsider.tech/2026/05/22/googles-ai-agent-ambitions-are-real-but-remain-locked-behind-a-100-monthly-paywall/ (T3, paywall and limitations critique)

Caveats: No Google primary announcement specifically for the May 26 broader rollout; inferred from the I/O timeline and corroborating reports. Google’s own in-product disclosure warns the agent may take sensitive actions without per-action confirmation — users should review permission scope before connecting financial or calendar accounts. EU/UK availability contingent on AI Act compliance; Q3 2026 is an analyst estimate, not a Google commitment.

Gemini 3.5 Flash becomes the global default model in Google AI Mode

Source: https://blog.google/products-and-platforms/products/search/search-io-2026/ · Google · 2026-05-19 (deployment effective 2026-05-26) Verification: T3 secondary · deployment milestone · model-release Tier nuance: The May 26 deployment date confirmed by search engine coverage and Google AI Mode tracker; no standalone Google announcement post for the rollout date specifically.

Google deployed Gemini 3.5 Flash as the default model powering AI Mode in Google Search on May 26, 2026, making it the AI engine for a feature that has reached one billion monthly users across approximately 200 countries and 98 languages. Gemini 3.5 Flash was announced at Google I/O on May 19 as the first in the Gemini 3.5 series; Google described it as delivering frontier-class performance at Flash-class latency, and as the engine behind the Gemini Enterprise Agent Platform managed agents released in public preview on May 19. The May 26 deployment also included expansion of Personal Intelligence in AI Mode — the ability to securely connect Google apps like Gmail and Google Photos — to users in approximately 200 countries.

Why it matters for automation/productivity: Gemini 3.5 Flash is now the inference backbone for a billion-user search experience and the foundation for Gemini API managed agents. Teams evaluating Gemini-based agent pipelines should run benchmarks against 3.5 Flash rather than 3.1 variants; the model is available in the Gemini API and Google AI Studio today.

Key claims:

• Deployment as AI Mode default: 2026-05-26 → multiple search-coverage outlets (T3)
• AI Mode monthly users: 1 billion as of May 19 → Google I/O announcement (T2)
• Personal Intelligence expansion: ~200 countries → Google Search I/O blog (T2)

Cross-references:

• https://ai.google.dev/gemini-api/docs/changelog (T2, no explicit May 26 entry; last entry was May 19 model GA)
• https://discoveredlabs.com/blog/google-ai-mode-may-2026-search-update (T3, AI Mode impact analysis)

Caveats: No standalone Google announcement post for the May 26 deployment date; date inferred from corroborating reporting. Independent benchmark coverage of Gemini 3.5 Flash is limited; one 5-day independent evaluation published around May 24 assigned it an Intelligence Index rank of #8 and flagged a 61% hallucination rate — weaker than Google’s I/O figures. Independent audited benchmarks are not yet available.

Anthropic appoints KiYoung Choi as Korea Representative Director ahead of Seoul office opening

Source: https://www.anthropic.com/news/kiyoung-choi-representative-director-anthropic-korea · Anthropic · 2026-05-26 Verification: T2 verified · announcement · ai-for-business

Anthropic appointed KiYoung Choi as Representative Director of Anthropic Korea on May 26, 2026, ahead of opening a Seoul office — its third Asia-Pacific location after Tokyo and Bengaluru. Choi joins from Snowflake, where he served as General Manager for Korea, and has held country-level roles at Google Cloud, Adobe, Autodesk, and Microsoft spanning 30+ years. Anthropic cited Korea as one of its five highest-usage markets globally, with Koreans using Claude at more than 3.5 times the expected rate relative to population size. The Korea team will focus on enterprise and startup partnerships, government and research institution engagement, and developer community support.

Why it matters for automation/productivity: For organizations in Korea evaluating Claude-based workflows, a local Anthropic presence typically accelerates enterprise contract cycles, provides in-region support escalation paths, and signals sustained product investment. The appointment confirms Anthropic’s continued APAC expansion beyond Japan and India.

Cross-references:

• https://www.anthropic.com/news/seoul-becomes-third-anthropic-office-in-asia-pacific (T2, October 2025 Seoul office announcement, providing background)
• https://finance.yahoo.com/news/anthropic-open-south-korea-office-120403852.html (T3, corroborating)

Suno AI raising $250M+ at $5B valuation led by Bond Capital

Source: https://www.axios.com/pro/all-deals/2026/05/26/suno-spotify-remix-bond-raise · Axios · 2026-05-26 Verification: T3 secondary · deal reporting · ai-for-business

Bond Capital is leading a funding round for AI music startup Suno that is expected to exceed $250 million and would value the company at approximately $5 billion, Axios reported on May 26, 2026. The round is not yet closed. Suno’s last financing in November 2025 raised $250 million at a $2.45 billion post-money valuation; the proposed new round would nearly double that valuation. Suno has ongoing copyright litigation from major record labels — filed in June 2024 and still unresolved — which remains a material risk factor for investors.

Why it matters for automation/productivity: Suno’s financing signals continued investor appetite for AI audio generation infrastructure at scale. For teams building content pipelines that incorporate AI-generated music or sound effects, Suno’s API — currently available at usage-based pricing — remains one of the few commercial options with both a licensing framework and a model capable of full-length music generation; Stability AI’s Stable Audio 3.0 (released May 20) is the primary open-weight alternative.

Key claims:

• Round size: $250M+ → Axios primary (T2, deal reporting)
• Valuation target: $5B → Axios (T2)
• Bond Capital as lead → Axios (T2)
• Prior round: $250M at $2.45B, November 2025 → TechCrunch (T2)

Cross-references:

• https://www.musicbusinessworldwide.com/suno-is-raising-another-250-million-at-a-5-billion-valuation-reports/ (T3, corroborating)
• https://www.digitalmusicnews.com/2026/05/04/suno-series-d-reports/ (T3, earlier reporting on the round)

Caveats: Round not yet closed as of May 26 reporting. Ongoing copyright litigation from major record labels is an unresolved risk. Deal terms may change.

Fireworks AI in funding talks targeting $15B valuation

Source: https://www.valuethemarkets.com/cryptocurrency/news/fireworks-ai-seeks-15-billion-valuation-in-new-funding-round · Value The Markets · 2026-05-27 Verification: T3 secondary · deal reporting · ai-for-business

Fireworks AI, a Redwood City-based AI inference platform, is in talks to raise a new funding round at a $15 billion valuation, according to reporting on May 27, 2026. The company was valued at $4 billion in October 2025 following a $250 million Series C led by Lightspeed Venture Partners, Index Ventures, and Evantic Capital, with NVIDIA and AMD participating. If the round closes at the reported figure, the valuation would represent a 3.75x increase in approximately seven months. The company operates a cloud inference platform for deploying and scaling AI models at enterprise scale, competing with AWS, Google Cloud, and Azure inference offerings.

Why it matters for automation/productivity: Fireworks AI provides a managed inference API that can run a wide range of open and proprietary models, including Llama 4, Mixtral, and custom fine-tunes, often at lower latency and cost than hyperscaler equivalents for high-throughput use cases. Rising valuation reflects investor demand for specialized inference infrastructure as model deployment costs become a significant budget line.

Key claims:

• Valuation target: $15B → Value The Markets (T3, citing Bloomberg sourcing)
• Previous Series C: $250M at $4B, October 2025 → Fireworks AI official blog (T2)

Cross-references:

• https://fireworks.ai/blog/series-c (T2, October 2025 Series C announcement)
• https://sacra.com/c/fireworks-ai/ (T3, company profile with funding history)

Caveats: Talks reported, no term sheet or close announced. Multiple outlets summarizing the same Bloomberg-sourced reporting; primary Bloomberg article not directly accessible. Round size not disclosed in available reporting.

Dropped

Items considered but not published, with reason.

Limitations

• Window: Strict 72h window (May 24–27) yielded 7 items. Prior bulletins (May 22–26) covered most of the week’s major announcements, especially from the Google I/O period (May 19–22). This bulletin reflects genuinely new signal from May 25–27.
• Sources unreachable: x.ai/news/grok-build-cli returned HTTP 403 in this run. Grok Build CLI item verified via CryptoBriefing, Engadget, and The Tech Outlook — consistent on date, features, and pricing. openai.com/news returned 403; Codex CLI release verified via GitHub releases (T2). Google AI Mode deployment date (May 26) has no standalone blog post; inferred from I/O announcement and corroborating media coverage.
• Login-walled coverage: X timelines, Instagram, LinkedIn, and Discord were not accessed directly. Searches for @AnthropicAI, @xai, @OpenAI, and @GoogleDeepMind via indexed results found no in-window product announcements beyond items already discovered.
• Gemini 3.5 Flash independent benchmarks: One 5-day independent evaluation published around May 24 reported an Intelligence Index rank of #8 and a 61% hallucination rate, weaker than Google’s I/O figures. No audited head-to-head evaluation across all Gemini 3.5 Flash use cases is available as of this run.
• Grok Build benchmarks absent: No independent benchmark (SWE-bench or equivalent) has tested the production Grok Build CLI model (grok-build-0.1). The 70.8% SWE-bench figure in circulation applies to the deprecated grok-code-fast-1.
• Suno and Fireworks funding rounds not confirmed closed: Both are deal-in-progress reports. Final terms, investor composition, and amounts may differ.
• SEA-region coverage: Searches for Indonesian and Southeast Asian AI product launches yielded no in-window items. Coverage this period is US-heavy.
• Categories with thin or no coverage: agent-framework, mcp-ecosystem, research-papers, productivity-ai (assistive), policy-regulation — no new items in the strict window.

Search log (compact)

Q: Anthropic Claude announcement May 2026 → 10 results, 5 high-relevance
Q: OpenAI announcement release May 26 27 2026 → 10 results, 3 high-relevance
Q: Google DeepMind Gemini release May 26 27 2026 → 10 results, 3 high-relevance
Q: Anthropic news page (fetch) → 2 in-window items found (KiYoung Choi, Chris Olah remarks)
Q: AI agent framework launch release May 25 26 27 2026 → 10 results, 2 high-relevance
Q: MCP Model Context Protocol new server update May 26 27 2026 → 10 results, 1 high-relevance
Q: Anthropic PwC and SAP partnerships (fetch) → May 14 and May 12 — outside window
Q: OpenAI Codex goals update May 26 2026 changelog → 10 results, 4 high-relevance
Q: OpenAI Codex CLI changelog (fetch) → 0.134.0 May 26 confirmed
Q: xAI Grok Meta Llama Mistral announcement May 26 27 2026 → 10 results, 2 high-relevance
Q: AI news launch release May 27 2026 → 10 results, 3 high-relevance
Q: Grok Build launch beta May 25 2026 xAI coding → 10 results, 8 high-relevance
Q: AI startup funding announcement product launch May 25 26 27 2026 [exploratory] → 10 results, 2 high-relevance
Q: Anthropic KiYoung Choi (fetch) → May 26 confirmed, details retrieved
Q: Grok Build Engadget and CryptoBriefing (fetch) → May 25 confirmed, features retrieved
Q: Google Deep Research Max autonomous research agent May 26 2026 → 10 results, April 21 date found
Q: SAP Anthropic partnership (fetch) → May 12, outside window
Q: new AI model release announcement May 26 27 2026 → 10 results, 1 high-relevance in window
Q: OpenAI ChatGPT new feature update May 26 27 2026 → 10 results, 2 high-relevance
Q: ChatGPT releasebot (fetch) → No updates for May 24–27 window
Q: GitHub trending AI repos stars week May 27 2026 [exploratory] → 10 results, 2 high-relevance
Q: GitHub trending page (fetch) → knowledge-work-plugins (Jan 30 launch confirmed), Understand-Anything trending
Q: Cursor Claude Code dev tools update May 26 27 2026 → 10 results, 1 high-relevance
Q: Claude Code releasebot (fetch) → Last update May 23 (covered in prior bulletin)
Q: Grok Build controversy criticism limitation May 2026 [adversarial] → 10 results, 3 high-relevance
Q: new LLM model open source released May 25 26 27 2026 → 10 results, 0 high-relevance in window
Q: Hugging Face Papers (fetch) → QUEST, CUA-Gym, MemForest (May 26 arXiv, T3)
Q: Stability AI Stable Audio 3 open weights API release May 26 2026 → 10 results, May 20 as primary date
Q: Suno AI $250 million funding Bond Capital May 26 2026 → Axios primary confirmed
Q: Fireworks AI $15 billion valuation funding talks date May 27 → ValueTheMarkets May 27 confirmed
Q: AI Indonesia OR startup AI Asia Tenggara OR model AI lokal May 2026 [SEA exploration] → 8 results, 0 in-window
Q: Gemini AI Mode global rollout May 26 2026 → 10 results, confirmed May 26 deployment
Q: Gemini Spark availability launch date user report May 26 27 2026 → 10 results, broader US week of May 26 confirmed
Q: Anthropic $900 billion round closed confirmed May 27 2026 [adversarial] → 10 results, no primary confirmation
Q: Google Gemini Spark problems limitations user complaints May 2026 [adversarial] → 10 results, paywall and privacy concerns surfaced
Q: Anthropic $900 billion round criticism concerns investors [adversarial] → 10 results, DOD litigation risk and early-backer skepticism noted

Total searches: 35, of which ~12 exploratory, adversarial, or SEA/lateral (35%). Primary source fetches: 8 (anthropic.com/news, Codex CLI GitHub releases, KiYoung Choi Anthropic, Grok Build CryptoBriefing, ChatGPT releasebot, GitHub trending, Hugging Face papers, ValueTheMarkets).

Suggested next runs

• Anthropic $30B round confirmation — round expected to close week of May 26; once Anthropic publishes a primary announcement, the T3 item in the May 26 bulletin upgrades to T2 verified with final valuation and investor names
• Grok Build independent benchmark — no SWE-bench or equivalent score exists for grok-build-0.1; first independent evaluation will clarify positioning versus Claude Code (87.6%) and Codex CLI (88.7%)
• Gemini Spark practitioner reports — first user reports on task completion accuracy, safety guardrail behavior, and EU rollout timeline expected within days of broader US access; worth a follow-up item when sufficient practitioner data accumulates
• Suno/Fireworks round closings — both rounds are in progress; confirmations or material changes in terms warrant brief follow-up items