AI News Highlights for May 29, 2026
- Claude Opus 4.8 is live: Anthropic shipped its new flagship the same day it was announced, with pricing held flat and agent benchmarks improving across the board.
- Anthropic overtakes OpenAI at a $965B valuation: The company closed a $65B Series H, and reports say the round — likely its last private raise before an IPO — pushed its valuation past OpenAI's.
- Apple is expected to overhaul Siri in iOS 27: Bloomberg reports it will be announced at WWDC on June 8, with Claude and Gemini said to be selectable as the default AI.
- Moves from the platforms: Microsoft added Mistral Medium 3.5 to Copilot Studio, and YouTube began auto-detecting AI-generated content to apply labels.
1. Anthropic Releases Claude Opus 4.8 — Flat Pricing, Stronger Agent Performance
On May 28, 2026, Anthropic announced Claude Opus 4.8, its new flagship model, and made it available the same day. Compared with the previous-generation Opus 4.7, it is reported to improve across coding, agentic (AI that completes tasks autonomously), and knowledge-work benchmarks, while pricing stays flat at $5 per million input tokens and $25 per million output tokens (Claude Opus 4.8 — Anthropic official).
New capabilities include "Effort Control" in claude.ai, which lets you dial how much effort the model spends on a response, and "Dynamic Workflows" in Claude Code, which handles large-scale code migrations using hundreds of parallel subagents. There is also a "Fast Mode" that runs roughly 2.5x faster and is about 3x cheaper than the previous generation's Fast Mode — note that this does not mean it is cheaper than standard mode (Anthropic official).
On behavior, Anthropic says the model "reports progress more honestly, flags uncertainty, and is less likely to make unsupported claims" (Axios). On benchmarks, it scored 84% on Online-Mind2Web, which measures web operation, beating both Opus 4.7 and GPT-5.5, and it became the first model to exceed 10% on the Legal Agent Benchmark under an all-pass criterion. It is also reported to improve on Terminal-Bench 2.1 and OSWorld-Verified (Gizmodo).
Item | Details |
|---|---|
Pricing (per million tokens) | $5 input / $25 output (unchanged from Opus 4.7) |
Online-Mind2Web | 84% (beats Opus 4.7 and GPT-5.5) |
Legal Agent Benchmark | First to exceed 10% on the all-pass criterion |
Key new features | Effort Control / Dynamic Workflows / Fast Mode (~2.5x faster, ~3x cheaper than the previous-gen Fast Mode) |
So what does this mean? The practical takeaway is that agent performance was raised without raising prices. In particular, Fast Mode and parallel subagents push down both the cost and the wait time of large-scale code migrations and repetitive processing, so teams weighing automation in development and knowledge work have good reason to revisit their cost estimates.
2. Anthropic Raises $65B at a $965B Valuation — Reported to Overtake OpenAI
On May 28, 2026, Anthropic closed a $65B Series H, and reports say its post-money valuation reached $965B. That is above OpenAI's private valuation as of late March (about $852B), and the round is positioned as what is likely its last private raise before an IPO (TechCrunch, Bloomberg).
The round is said to have been co-led by Altimeter, Dragoneer, Greenoaks, Sequoia, Capital Group, Coatue, and D1, of which $15B is reported to be previously committed funding from hyperscalers (including the $5B from Amazon announced in April). Run-rate revenue (the annualized pace of revenue) crossed $47B this month, a sharp rise from $30B at the start of the year and $10B in annual revenue the year before. The Wall Street Journal reports 130% growth and an outlook for the company's first operating profit (CNBC).
This is the confirmation and upward revision of the initial "~$900B valuation" report we covered in our May 25 article.
So what does this mean? More important than the valuation overtaking itself is that the surge in run-rate revenue and the outlook for an operating profit — the "business fundamentals" behind it — are being discussed at the same time. With both funding and revenue starting to turn, that can be reassuring for business continuity when you embed Claude into your operations, though it is worth remembering these figures are based on reporting.
3. Apple Reported to Fully Revamp Siri in iOS 27 — Expected at WWDC on June 8
According to a report by Bloomberg's Mark Gurman, the centerpiece of iOS 27 is expected to be a full overhaul of Siri. Cited features include a permanent presence in the Dynamic Island, a ChatGPT-style standalone app, and camera integration, and it is reported that Siri will be able to check free time, flag scheduling conflicts, and draft emails, notes, and messages using information from the web and on the device (Bloomberg).
Also said to be planned are AI-generated wallpapers, a system-wide grammar checker, a revamped Image Playground, and new Photos AI tools called "Reframe" and "Extend." Furthermore, on iOS, iPadOS, and macOS 27, it is reported that Apple Intelligence's default AI will be selectable as something other than ChatGPT (such as Claude or Gemini) (MacRumors).
So what does this mean? If the reports hold, the big deal is that hundreds of millions of iPhones would let you choose Claude or Gemini as the default AI. This is a move toward making the OS-standard assistant experience multi-model, which could expand the distribution channels available to each AI vendor. That said, this is not a confirmed announcement at this point — we are waiting for the official reveal at WWDC on June 8.
4. Microsoft Adds Mistral Medium 3.5 to Copilot Studio
On May 28, 2026, Microsoft announced that it had added Mistral Medium 3.5 to the model lineup of Copilot Studio, its agent-building platform. It is available worldwide in the early-release environment, but it is positioned as experimental and not recommended for production use (Microsoft official).
Mistral Medium 3.5 is said to be well suited to long-running tasks, reliable calls to multiple tools, and structured output, and it lets you set the reasoning effort (the amount of thinking spent on a response) per request. For organizations in the EU, it is pitched on the benefit of in-region data processing. Using it requires a two-step opt-in across the M365 admin center and the Power Platform admin center (Microsoft official).
So what does this mean? With more model options available in Copilot Studio, it becomes easier to use different models depending on the use case and data residency. For organizations in the EU that prioritize where data is processed in particular, this widens the range of choices. Because it is offered as experimental for now, however, production use calls for careful judgment.
5. YouTube Begins Auto-Detecting AI-Generated Content for Labels
On May 27, 2026, YouTube announced that it had moved disclosure labels for photorealistic (looking as real as a photo) or significantly altered AI content to a more prominent position. On long-form videos they appear directly below the player, and on Shorts they appear as an on-screen text overlay (YouTube official).
In addition, YouTube started auto-detection that automatically applies a label when the system detects "significant photorealistic AI use" even if the creator has not disclosed it. False positives can be disputed in YouTube Studio, and disclosures from in-house tools (Veo / Dream Screen) or C2PA metadata are kept permanently. Notably, it is stated that "the label itself does not change recommendations or monetization eligibility" (YouTube official).
So what does this mean? For creators who use AI to produce videos, the practical point is that disclosure has taken a step forward from "left to self-reporting" to "self-reporting plus auto-detection." It is said not to affect monetization or recommendations, but knowing the dispute procedure for false positives offers peace of mind.
Unconfirmed but Worth Watching: This Week's AI Leaks and Rumors (Not Confirmed Information)
Note: the following is unconfirmed information. Please read it as "developments worth watching" until primary sources confirm them.
A. Anthropic's "Mythos-class" Model May Reach General Availability Within Weeks
Confidence: High / Sources: Anthropic official / Axios
Alongside the Opus 4.8 announcement, Anthropic referenced a higher-end "Mythos-class" model, and Axios notes that availability could come "within weeks." Right now, only about 50 partners in Project Glasswing are using "Claude Mythos Preview" for cybersecurity purposes, and it is reported to have found more than 10,000 high- and critical-severity vulnerabilities in one month (Anthropic official). Access may broaden on the condition that stronger safeguards are put in place, but the exact date, the scope of general availability (GA), and pricing remain unconfirmed (Axios).
B. OpenAI May Be Preparing "Remote Codex Control" for the ChatGPT Mobile App
Confidence: Medium / Source: TestingCatalog
According to TestingCatalog, a feature that lets you remotely control Codex on a connected Mac from the ChatGPT mobile app appears to be in the works. It is reported that operations such as starting and continuing threads, approving actions, and reviewing diffs and test results are envisioned. A "Personal Finance (household dashboard)" for Pro users may also be rolling out in the United States. The availability scope and timing of a full rollout for both appear to be unconfirmed (TestingCatalog).
C. The "Real GPT-6" Is Said to Be Expected in Q3–Q4 2026
Confidence: Medium-Low / Sources: Leakers such as Jimmy Apples (@apples_jimmy) plus Polymarket
The model rumored in April to be "GPT-6" appears to have landed as GPT-5.5 (codename Spud). The real GPT-6 is said to be expected in the second half of 2026 (Q3–Q4) as the base scenario, and there are observations that prediction markets also see a within-year release as likely. That said, there is no official confirmation from OpenAI, and this should be read strictly as a possibility based on leaker and prediction-market observations (adam.holter.com).
Summary
The AI landscape as of May 29 was defined by the evolution of the models themselves (Claude Opus 4.8) moving in step with the funding and revenue behind them (Anthropic's $965B valuation and surging revenue). With both performance and business foundations advancing, the assumptions for anyone weighing AI adoption in their operations are being updated.
Meanwhile, Apple's iOS 27 Siri overhaul is still at the reporting stage, and the official reveal at WWDC on June 8 is the next thing to watch. Platform-side groundwork is also progressing — Microsoft's model lineup and YouTube's auto-labeling among them — adding more practical questions around "which model, where, and with what disclosure to use it." Separating confirmed information from observation and making decisions that fit your own use case is what matters.