Anthropic Launches Claude Opus 4.8 and Completes $65B H-Round Funding

As IPO expectations heat up, Anthropic is accelerating the pace of its product models.

Just now, Anthropic announced two major updates: first, upgrading its flagship model to Claude Opus 4.8, and second, completing a $65 billion Series H funding round, bringing its post-money valuation to $965 billion, nearing the $1 trillion mark.

For an AI company whose valuation is nearing $1 trillion, the market is no longer just looking at model benchmarks—it’s watching whether the company can turn intelligence, tools, development environments, cloud platforms, and computing resources into a scalable, deliverable infrastructure.

In other words, Anthropic must gradually shift from being a company that delivers good models to one that aims to reshape how enterprises use AI.

The newly released Claude Opus 4.8 is an upgrade to Anthropic’s flagship Opus series. Given the current pace of model releases, as netizens jokingly put it, we’ll likely see Claude Opus 6 before GTA 6 launches (assuming no delays, in November).

Anthropic states that Opus 4.8 builds upon Opus 4.7 with improvements in coding, agent tasks, reasoning, and knowledge work, and is now available to users, with standard pricing unchanged at $5 per million input tokens and $25 per million output tokens.

Developers can also use claude-opus-4-8 via the Claude API.

API price comparison https://platform.claude.com/docs/en/about-claude/models/overview

According to officially disclosed information, the improvements in Opus 4.8 cover coding, agent capabilities, reasoning ability, and practical knowledge-based tasks.

Anthropic included a comparison chart in its release materials showing Opus 4.8’s performance against its predecessor, Opus 4.7, and other models across multiple tests. Of course, the primary focus of the model’s upgrade lies in its performance on long tasks and complex collaborations, rather than single-response quality.

In real-world workflows, models often need to sequentially handle multi-step tasks, invoke tools, evaluate intermediate results, and proceed based on feedback. Anthropic states that early testers found Opus 4.8 to be more reliable and clearer in its judgments when executing agent tasks.

The improvement in honesty with Opus 4.8 is a major highlight.

A common issue with AI models is making premature judgments with insufficient evidence and confidently claiming progress. Anthropic states that Opus 4.8 is more likely to acknowledge uncertainty in its work and makes fewer unsupported claims.

Code tasks particularly reflect this change.

Internal evaluations show that Opus 4.8 is approximately one-fourth as likely as its predecessor to generate code defects that go unreported. This means the new model is more likely to alert users to potential risks rather than letting issues slip through to later testing or production environments.

In terms of alignment and safety, Anthropic has stayed true to its core narrative. Opus 4.8 shows a significantly lower rate of misaligned behaviors—such as deception and enabling abuse—compared to Opus 4.7, and is now close to the alignment performance of one of the best-performing models currently available, Claude Mythos Preview.

Security, reliability, and controllability remain the key terms Anthropic uses to differentiate itself. As Claude becomes more deeply integrated into enterprise workflows, these terms are taking on greater business significance.

Interestingly, after Opus 4.8 was released, netizens noticed something was off.

https://x.com/realNyarime/status/2060059543820963975

Many netizens who tested found that when they asked Opus 4.8 about its model identity, it did not always respond with "Claude."

Sometimes it identifies itself as Qwen, and other times it claims to be DeepSeek, suggesting possible distillation behavior.

When users ask the same question on Claude’s official client, such responses are often harder to reproduce, likely because the system prompts and product-level constraints within the client are more comprehensive.

Alongside the launch of Claude Opus 4.8, several new products and developer features are now available.

Among these, the most direct impact on Claude users' experience is effort control, also known as thought intensity adjustment.

The control is located next to the model selector; as the name suggests, users can decide how much reasoning power Claude should dedicate to a single task. At higher intensities, Claude performs more reasoning to deliver higher-quality responses; at lower intensities, Claude responds faster and consumes credits more slowly.

Anthropic states that Opus 4.8 defaults to high effort; users can also select extra, which corresponds to xhigh in Claude Code, or choose max to allow the model to use more tokens. Anthropic recommends that difficult tasks and long-running asynchronous workflows are better suited for extra.

What truly shapes the product form of Claude Code is dynamic workflows.

This feature is currently in research preview, designed to enable Claude Code to handle large-scale tasks that previously required lengthy engineering cycles. Work that once took quarterly planning can now potentially be completed in just a few days.

The core mechanism of dynamic workflows is that Claude dynamically generates orchestration scripts based on user tasks and runs dozens to hundreds of parallel subagents within a single session. The model first plans the task, then assigns it to multiple subagents, reviews the returned results, and finally reports back to the user. With the launch of Opus 4.8, these agents can now operate for longer durations.

This feature is primarily designed for complex, large-scale, or legacy codebases. Typical use cases include identifying bugs across the entire service, performance optimization audits, security audits, large-scale codebase migrations, framework replacements, API deprecation migrations, language porting, and multi-angle validation of critical solutions.

For usage, Anthropic recommends enabling auto mode in dynamic workflows. Users can directly ask Claude to create a workflow, or enable Ultracode in Claude Code. Ultracode sets the thought intensity to xhigh and allows Claude to automatically determine whether the current task is suitable for using a workflow.

Dynamic workflows are now available in the Claude Code CLI, Desktop, and VS Code extension for Max, Team, and Enterprise plans. By default, Dynamic workflows are disabled for Enterprise at launch and must be enabled by administrators in the Claude Code settings.

This feature is also available for Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry. For Max and Team users, as well as those using Claude Code via API, dynamic workflows are enabled by default.

Anthropic's migration case with Bun demonstrates the upper limits of dynamic workflows. Jarred Sumner used this feature to port Bun from Zig to Rust, generating approximately 750,000 lines of Rust code, achieving a 99.8% pass rate for the existing test suite, and completing the process from first commit to merge in about 11 days.

The entire migration process consists of multiple workflows: first, mapping struct fields in the Zig codebase to Rust lifetimes, then generating consistent .rs files for each .zig file, with hundreds of agents working in parallel and each file reviewed by two reviewers. Afterward, a fix loop continuously runs the build and test suite until both pass. Once migration is complete, an overnight workflow addresses unnecessary data duplication issues and opens PRs for each category of issue, ready for final review.

In addition to Claude Code, Anthropic has updated the Messages API. The Messages API now accepts system entries within the messages array.

Developers can update Claude’s instructions during task execution without disrupting the prompt cache or requiring updates to be passed through a user turn. This capability enables runtime updates to permissions, token budgets, or environmental context for agents.

Next, Anthropic plans to introduce a new model class even more intelligent than Opus—yes, the incredibly powerful Claude Mythos Preview. Mythos-class models are expected to be available to all customers within the coming weeks.

At that time, we will also be the first to taste the flavor.

On the same day as the release of Claude Opus 4.8, Anthropic announced the completion of a $65 billion Series H funding round, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, bringing its post-money valuation to $965 billion.

This round also includes $15 billion in existing commitments from hyperscalers, including $5 billion from Amazon. Strategic infrastructure partners such as Micron, Samsung, and SK hynix have also joined. Anthropic stated that these companies play a critical role in global memory, storage, and logic chip supply, enabling it to scale computing capacity as demand for Claude grows.

The expansion of computing power is the key backdrop behind this funding round. Anthropic has disclosed several infrastructure agreements: a deal with Amazon for up to 5 gigawatts of additional capacity; agreements with Google and Broadcom for 5 gigawatts of next-generation TPU capacity; and a partnership with SpaceX for access to GPU capacity in Colossus 1 and Colossus 2.

Anthropic also emphasized that Claude is the first frontier model to be available simultaneously on AWS, Google Cloud, and Microsoft Azure. However, AWS remains Anthropic’s primary cloud service provider and training partner.

Behind the financing lies a shift in Anthropic’s business positioning. Early on, large model companies competed on model capability and general chat experience; now, enterprise customers are more concerned with whether AI can be integrated into core workflows, handle complex tasks, and connect with development environments, cloud platforms, and internal systems.

Claude Code, Cowork, effort control, dynamic workflows, and the Messages API updates are all centered around this direction.

When viewed together, the product launch and funding round indicate that Anthropic is simultaneously expanding three key capabilities.

First, model capabilities: Opus 4.8 enhances performance in coding, reasoning, agent tasks, and knowledge work, while improving the expression of uncertainty.

Second is workflow capability: dynamic workflows enable Claude Code to move beyond single-code assistance to more complex engineering execution and review.

Third is infrastructure capability, supported by $65 billion in funding, commitments from hyperscale cloud providers, partnerships with memory and chip suppliers, and computing power agreements with Amazon, Google, Broadcom, and SpaceX to meet future model training and inference demands.

This is also the core logic behind Anthropic’s valuation approaching one trillion dollars. Claude is no longer just an AI chat interface—it is evolving into a workflow system that connects models, code, enterprise processes, cloud platforms, and computing infrastructure.

Opus 4.8 is the latest foundational model in this system; dynamic workflows represent the product form designed for complex engineering tasks; and the $65 billion in funding and expanded computing power are prerequisites for scaling this system to serve larger customer bases.

The tide of AI has propelled Anthropic to the forefront; from this height, forward means riding the wind, backward means sinking into the water—there is no third stance.

This article is from the WeChat official account "APPSO," authored by APPSO, discovering tomorrow's products.