Anthropic closes Series H at $65B / $965B post-money — Claude Opus 4.8 ships the same day

What’s confirmed

The reporting from Anthropic’s official Series H announcement, TechCrunch, CNBC, Sherwood News, and Wall Street Journal corroborates:

• Total raised: $65 billion
• Post-money valuation: $965 billion (just below the $1T threshold)
• Pre-money valuation: $900 billion (consistent with the agreed-terms reporting earlier in May)
• Round size escalated: from the initially reported $30B to the closed $65B
• Round co-leads: Altimeter Capital, Dragoneer, Greenoaks, Sequoia Capital, Capital Group, Coatue, D1 Capital Partners
• Strategic memory-chip partners: Samsung, SK Hynix, Micron — joining cap table for the first time
• Run-rate revenue: $47 billion (up from the $30B run-rate disclosed in April)
• Path to profitability: Wall Street Journal reports 130% revenue surge expected to deliver first operating profit, building on the Q2 projected $559M profitable quarter

The valuation in context

Anthropic now sits $113 billion above OpenAI at the closed-round comparison — having been substantially below OpenAI’s valuation as recently as February 2026. The trajectory: $61.5B (early 2025) → $183B (March 2025) → $350B (February 2026) → $965B (today). That’s a 15.7× jump in 16 months.

The memory-chip strategic investment angle

The most interesting unreported angle: Samsung, SK Hynix, and Micron joined the round. These are the world’s three largest HBM (high-bandwidth memory) manufacturers — together supplying the memory that goes inside every NVIDIA H100, B100, B200, and Rubin chip that powers AI training and inference today.

That’s structurally different from a “blue-chip tech investor joining” story. Anthropic’s memory supply chain just acquired direct equity exposure to Anthropic’s growth. For the three memory makers, it’s a hedge — they want to be aligned with whoever wins the AI training-compute race, since their HBM demand depends on AI capex sustaining. For Anthropic, it secures preferential allocation when memory supply tightens (which it does, periodically, as new model generations launch).

This signals Anthropic expects to be at the front of the queue for next-generation HBM3e+, HBM4, and any post-HBM4 memory products. That’s a real competitive asset, not just balance-sheet padding.

The Claude Opus 4.8 release

Anthropic shipped Claude Opus 4.8 the same day as the funding announcement — a deliberate combined narrative.

Benchmarks vs. competitors

• SWE-bench Verified: 88.6% (vs Opus 4.7: 87.6%, GPT-5.5: 81%+, Gemini 3.1 Pro: 78%)
• OSWorld-Verified (agentic computer use): 83.4% (vs Opus 4.7: 82.8%, GPT-5.5: 78.7%, Gemini 3.1 Pro: 76.2%)
• Agentic coding: 69.2% (vs Opus 4.7: 64.3%, GPT-5.5: 58.6%, next competitor: 54.2%)
• Terminal-Bench 2.1: 74.6%
• GDPval-AA: 1890 Elo

The agentic coding gain (+4.9 points absolute, +7.6% relative) is the headline capability improvement. For Claude Code users running long-horizon agent tasks, this is a meaningful step up.

Honesty and reliability improvements

The more interesting framing — and the one Anthropic is leaning into for the launch — is on reliability:

• 4× less likely to miss code flaws than Opus 4.7
• First Claude model to score 0% on the “uncritically reporting flawed results” evaluation
• 10×+ reduction in overconfidence vs Opus 4.7
• Fails to raise important events to the user only 3.7% of the time

This is the “more honest Claude” framing several outlets are using. For production deployment where false-positive results have downstream consequences (security analysis, financial modeling, medical research), the reliability gain matters more than the capability headline.

Dynamic Workflows in Claude Code

The new Dynamic Workflows feature — currently in research preview — lets Claude write orchestration scripts that spawn tens to hundreds of parallel subagents in a single session, with iterative verification and resumable state.

This is Anthropic’s answer to Antigravity’s multi-agent orchestration — Google’s headline differentiation feature at I/O 2026. Now Claude Code matches that capability in research preview, with general availability expected in coming weeks.

Pricing unchanged

Critically, Opus 4.8 pricing is unchanged from Opus 4.7 — same $5/$25 per million input/output tokens. Anthropic is removing the upgrade-evaluation hurdle for teams already on the Opus rate card. Combined with the agentic capability lift and the reliability improvements, the upgrade decision is mostly a no-brainer for existing Opus users.

Availability

Claude Opus 4.8 is live on:

• Claude API
• Amazon Bedrock
• Google Cloud Vertex AI
• Microsoft Foundry

That cross-cloud availability matches Opus 4.7’s reach.

What it means for the broader AI market

For OpenAI: The structural pressure intensifies. Anthropic’s confidential S-1 will follow OpenAI’s by months, not years, and both companies will hit public markets in late 2026 to early 2027. The investor positioning for both IPOs is now actively forming. Anthropic’s narrative — “fastest-growing AI revenue + safety-leadership credentials + profitability path + most-valuable-private-AI-company” — is harder for OpenAI to counter than it was three months ago.

For Google: Antigravity 2.0’s multi-agent orchestration differentiation now has direct competition from Claude Code’s Dynamic Workflows research preview. The “Google is the only one with multi-agent” framing won’t hold past Q3 2026 if the Anthropic feature ships GA on schedule.

For Claude Code users: Dynamic Workflows is the most significant new capability since Claude Code launched. The “tens to hundreds of parallel subagents” framing — if it works as advertised in production — is the kind of capability that materially shifts what a single developer can accomplish in a day.

For Claude chat users: Opus 4.8 is now the default flagship model for Pro and Max subscribers. Quality should improve immediately; pricing doesn’t change.

For the broader AI Harnesses category: Claude Code’s lead now extends across capability (Opus 4.8), orchestration (Dynamic Workflows), and infrastructure (the Series H capital base). The harness category’s positioning ladder is becoming substantially harder to climb for anyone outside the top tier.

The honest caveats

Three caveats worth surfacing:

Series H is not the IPO. “Last private round before IPO” is plausible but not guaranteed. AI valuations can shift between final private round and pricing day. Treat the $965B as the close valuation, not the public-markets pricing.

$47B run-rate is not annual revenue. Run-rate annualizes current-month revenue — useful as a leading indicator but not a substitute for booked GAAP revenue. The actual 2025-2026 calendar year revenue figures will be lower.

Opus 4.8 benchmarks are Anthropic’s own results. Independent benchmark verification will land over the coming weeks. The 88.6% SWE-bench Verified figure should be treated as Anthropic’s reported result pending independent verification.

What it changes for Pick Right readers tomorrow

If you’re a Claude Pro or Max subscriber, Claude Opus 4.8 is now the default Opus model for you. Quality should improve immediately on hard reasoning tasks. Pricing doesn’t change.

If you’re a Claude Code user, Dynamic Workflows is worth signing up for the research preview — particularly if you have any task that decomposes naturally into parallel work.

If you’re following Anthropic vs OpenAI as a competitive positioning story, this is the moment that crystallized: Anthropic is now the leader, not the challenger, by the most concrete metric the AI industry has — investor consensus on valuation. The Q3-Q4 2026 IPO race will be one of the most-watched in tech history, and both companies are positioning aggressively.

For broader context, see the Claude review, the Claude Code review, the Claude vs ChatGPT comparison, the Claude Code vs Antigravity comparison, and the AI Harnesses category roundup for the broader 2026 positioning picture.

Questions or corrections? Email Pick Right. Want the full list? See all news.