Claude double king bomb! Officials Announce $44.07 Billion in Funding, Opus 4.8, the Most Powerful General Purpose Model, Debuts

Early this morning.AnthropicStrongest generic modelClaude Opus 4.8Officially released, the new model benchmark test comprehensively beyond the Gemini 3.1 Pro, Opus 4.7, only one inferior to the GPT-5.5, but its standard mode price remains unchanged, the fast mode price is only Opus 4.7's1/3.

At the same time, Anthropic also made an official announcement of aUS$65 billion (approximately RMB 440.694 billion)Wheel H MegafinancingThe post-investment valuation shot upUS$965 billion (about RMB 6.54 trillion)The total amount of the program has almost doubled from $380 billion (roughly Rs. 2.58 trillion) in February of this year.3 times. The round was led by a number of high-profile investments, including Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital.

Anthropic's latest valuation is about 1.13 times that of OpenAI. Last week, the Financial Times reported that OpenAI's latest valuation was $852 billion (about Rs. 5.78 trillion).

The biggest upgrade to Opus 4.8 is the dynamic workflow, which makes it possible to generate hundreds of parallel sub-intelligences to perform large-scale tasks such as large-scale codebase migrations of hundreds of thousands of lines of code from startup to merge.

Claude Opus 4.8 is now fully online, with standard mode pricing unchanged and fast mode at 1/3 the cost of Opus 4.7 and 4.6. Standard mode is priced at $5 per million incoming tokens (~RMB33.9) and $25 per million outgoing tokens (~RMB169.5), while fast mode is priced at $10 per million incoming tokens (~RMB67.8) and $50 per million outgoing tokens (~RMB338.9). Rs. 67.8) per million input tokens and $50 (Rs. 338.9) per million output tokens, and the fast mode is priced at $10 (Rs. 67.8) per million input tokens. Developers can use claude-opus-4-8 via the Claude API.

AI data platform Databricks reports that Opus 4.8 achieves “a quantum leap in reasoning power” in its Genie Data Intelligence, and due to its multimodal efficiencies in processing PDFs and charts, its “token cost is 611 TP4T lower than Opus 4.7 by 61%”.

Regarding future plans, Anthropic revealed in a blog post that they plan to launch a new series of models with a level of intelligence beyond Opus, and expect to open up the Mythos-level models to all customers in the coming weeks, with development of related protection solutions moving forward at a rapid pace.

After Claude's update, Musk even reposted a message praising them for “a job well done”, only to be trolled by commenters who said, "Why don't you go fix Grok instead of applauding someone else.

First, the network evaluation of visual quality improvement is obvious, but can not count the hand index

Anthropic has not officially released Opus 4.8 related cases, and many users have released their own usage on the social media platform X.

Some developers have compared the results of Opus 4.7, Opus 4.8 and Opus 4.8 Max using the Four Elements project, and the improvement in the visual quality of the model generation is obvious.

Early adopters of Opus 4.8 have rated it highly, saying that the new model could well be called Opus 5, and they have generated the following web page based on the new model.

A developer tested Opus 4.8 by running My World related tasks through Maximum Thinking Power (Max) mode. He commented that Opus 4.8 overall performance is good, compared to Opus 4.7 there is a significant improvement, all tasks in a single run all successful, but the ability to leapfrog breakthrough has not been achieved. The total cost of his calls was $9 to $10 (about Rs. 61 to Rs. 68).

Another developer tested the Lava Lamp Project, and his results showed that Opus 4.7 looked better than Opus 4.8.

Another user trolled for Opus 4.8 by uploading a picture of a palm with six fingers, asking Opus 4.8 how many fingers this picture had.Opus 4.8 replied that it was a normal human palm with five fingers.

A user on Reddit has already begun to relentlessly troll, “Opus 4.8 has seen a precipitous drop in performance since its release.” He analyzed that Anthropic might have started running the service with a 2-bit quantized model to save money.

Anthropic also admits in its blog that Opus 4.8 is not a big upgrade from its predecessor, but the improvements are all very tangible.

Second, the benchmark test almost fully exceeds the GPT-5.5, Gemini 3.1 Pro

In terms of benchmark performance, compared to Claude Opus 4.7, GPT-5.5, and Gemini 3.1 Pro, Opus 4.8 outperforms the other models across the board in terms of intelligentsia code development, cross-disciplinary reasoning, computer use, financial analysis of intelligentsia, and the ability to do work on high-value knowledge jobs, and is only marginally inferior in terms of intelligentsia endpoint programming to GPT -5.5.

Anthropic has released a system card for Opus 4.8, which mentions that Opus 4.8 is very close to Mythos, which is still in selective release, in terms of inconsistency behavior, with the deviation between the two being about 1.9, lower than the 2.5 of Opus 4.7 vs. Mythos.

The training data for Claude Opus 4.8 is Anthropic's proprietary hybrid dataset, which includes publicly available information from the Internet, public and private datasets, and synthetic data generated by other models, which it cleanses and filters by de-duplication and classification.

Its blog mentions that the strength of Opus 4.8 is that it is best suited to handle tasks and performance-demanding scenarios that could not be accomplished with the previous generation of models, and is designed for specialized software engineering, complex intelligentsia workflows, and high-risk enterprise-level tasks.

Opus 4.8 has an adaptive thinking feature that automatically adjusts its thinking inputs according to the complexity of the task, spending more time on difficult problems and responding quickly to simple ones, and its common application scenarios include:

Advanced programming, with minimal supervision, Opus 4.8 is able to deliver code that can be used in a production environment. The model is capable of thorough planning, remains stable for longer periods of time in continuous operation, and runs in large code bases. It also automatically detects and corrects errors.

AI intelligences, Opus 4.8 is capable of thoughtful planning, utilizing memory capabilities to learn across sessions, and driving jobs that require long runs with minimal human intervention.

Enterprise Workflow: On the enterprise side, the model delivers context across sessions with powerful spreadsheet, slide and document performance to manage complex, multi-day projects end-to-end.

Third, the user can manually control the intensity of thinking, more honest than the previous generation of models

Anthropic has also released several updates:

The first is dynamic workflow, based on which Claude can plan work and then run hundreds of parallel sub-intelligences in a single session, after which the output is validated and the results are fed back to the user. For example, Claude Code with Opus 4.8 is now able to perform large-scale codebase migrations of hundreds of thousands of lines of code from startup to merge, using an existing test suite as a baseline.

The feature is currently in research preview and can be experienced by users of Claude Code's Enterprise, Team and Max editions.

The second is the addition of Effort control for claude.ai and Cowork. Next to the model selector, Anthropic has added a button for users to set how much “effort” Claude puts into generating responses. With a higher Effort setting, Claude will think more often and more deeply, resulting in better responses, and conversely, he will think less and respond faster.

This feature is now available for all packages and users are free to choose.

Finally, the Claude Messages API now supports the inclusion of system commands within message arrays. Developers can dynamically update model guidelines during task execution without disrupting the cue word cache or having to resort to user turns to deliver updates. This capability enables developers to adjust permissions, token quotas, and environmental contexts in real-time while the intelligence is running.

Opus 4.8's honesty, early testers say Opus 4.8 is more likely to flag uncertainty in its work and less likely to make unsubstantiated assertions.Anthropic's internal evaluations show that the probability that a vulnerability in code written by Opus 4.8 letting itself go unchecked and undetected has been reduced to about one in four, as compared to its predecessor.

Fourth, the new financing contains $5 billion invested by Amazon

Officially announcing the funding, Anthropic also revealed that its annualized revenues had crossed $47 billion (roughly Rs. 318.6 billion) earlier this month. The financing is expected to be used to advance security and interpretability research, expand computing power to meet growing Claude demand, and expand the products and partnerships that customers rely on.

In addition to the previously mentioned lead investor, significant other investors in this round include AMP PBC, Baillie Gifford, Temasek, and more than a dozen other investment organizations.

It's worth noting that the round also covers the $15 billion (roughly Rs. 101.7 billion) in investments that the mega tech firms had previously promised to put in, which includes a $5 billion (roughly Rs. 33.9 billion) investment from Amazon.

Micron, Samsung and SK Hynix are synchronized as Anthropic's strategic infrastructure partners.

In recent weeks, Anthropic has dramatically increased its computing power, signing an agreement with Amazon to add up to 5 gigawatts of new computing power; agreements with Google and Broadcom to add 5 gigawatts of next-generation TPU computing power; and an agreement with SpaceX to use the GPU computing power of the Colossus 1 and Colossus 2.

Claude is also the world's first cutting-edge model to support all three of the world's major cloud platforms, Amazon Cloud Services, Google Cloud and Microsoft Azure, simultaneously. However, AWS remains its primary cloud service provider and training partner.

Conclusion: Anthropic under pressure as new model releases pick up pace?

As you can see, instead of pursuing disruptive capability upgrades, Claude's update aims to reduce costs.Anthropic has also revealed that they are continuing to develop and release multiple models that seek to achieve similar capabilities to the Opus line at a lower cost.

Opus 4.8 was released at a faster pace than before, coming only 41 days after Opus 4.7, compared to the previous three-month and seven-month release cycles for the Sonnet and Haiku series of models. Perhaps this is because the successive releases of major new versions of OpenAI's Codex and Google's Gemini Flash models have put pressure on Anthropic to keep up.

Article source: Wisdom