25 Latest AI News: Claude Opus 5, Midjourney V8.2, Grok and the Model Race

Bài viết do Ban Biên Tập Marketing365 thực hiện, biên tập theo Chính sách biên tập của Marketing365. Cập nhật lần cuối 25/07/2026.

Nội dung

A $10 million donation to protect Moon Camp tied to OpenAI leadership’s family
Claude system prompt leak heats up the debate on AI transparency
Midjourney V8.2 becomes the default, prioritizing aesthetics and personalization
Claude Opus 5 appears on Perplexity
New benchmark puts Claude Opus 5 at the top in knowledge work
Anthropic emphasizes security safeguards for Opus 5
ChatGPT web lets users share customizable “pets”
Palantir continues telling the story of unified data and “digital twins”
Claude Opus 5 ranks No. 1 on SWE-Bench, according to Cline
Grok speeds up its release cycle with 4.6 and 4.7 milestones
The SpaceX – Tesla – xAI scenario continues to fuel community discussion
Grok Odyssey is expected to launch this year
Anthropic is seen as stronger in many real-world task tests
Opus 5 is said to outperform the standard on ARC-AGI-3
The open source story continues to be linked to Elon Musk’s image
Anthropic is putting direct pressure on premium model pricing
Opus 5 is described as a major step forward in agentic coding
Anthropic launches Opus 5 with a near-top but cheaper message
Claude Opus 5 officially joins Anthropic’s new model lineup
Opus 5 is described as near frontier but with significantly lower per-task cost
The open source debate continues to fuel speculation
Anthropic positions Opus 5 as close to Fable 5 but cheaper
ChatGPT Work agent supports websites that require login
Cursor integrates Claude Opus 5 for coding, with zero data retention support
Midjourney acquires Co-Star, expanding its product design ambitions
What this means for the Vietnamese market
References

The global AI race continues to accelerate, with updates spanning language models, work tools, image generation, and enterprise applications. For Vietnamese marketers, this is not just a technology story, but a signal of how content, creativity, advertising, data analysis, and automation will change in the coming months.

What stands out in this round of news is that many platforms are shifting from “strong in demos” to “strong in real-world applications,” especially in coding, agentic workflows, personalization, and the ability to connect internal data. Below are the 25 latest developments to watch.

Key points:

Anthropic launched Claude Opus 5, positioned close to the leading model group but at a significantly lower cost according to published benchmarks.
OpenAI, Perplexity, Cursor, and Midjourney all rolled out product updates showing that AI is moving deeper into everyday work.
The competition is no longer just about being “smarter,” but also about being “cheaper, safer, and truly usable” in business and content creation.
For the Vietnamese market, now is the right time to review marketing workflows, test agents, and prepare a multi-model content strategy.

A $10 million donation to protect Moon Camp tied to OpenAI leadership’s family

Friends of Big Bear Valley said they raised the full $10 million needed to preserve Moon Camp, a 63-acre forest and water area in Big Bear. The community campaign aims to block a luxury housing and marina development project, as the area is home to the eagle pair Jackie and Shadow. Notably, Anna and Greg Brockman contributed more than $5.5 million; Greg Brockman is president of OpenAI.

From a communications perspective, this is an example of how senior AI industry figures are increasingly appearing in public stories beyond product news. For technology brands, personal credibility and social impact can indirectly shape corporate image.

Claude system prompt leak heats up the debate on AI transparency

A viral post claimed that the system prompt for Claude Opus 5 is very large and contains many new instruction blocks related to safety. While social media claims should be treated cautiously, the community reaction shows growing interest in how models are “controlled” from within.

For marketers and content teams, this story is a reminder that AI is not just the user interface people see, but also the layer of rules, constraints, and behavior design behind it. Transparency and quality control will be competitive advantages in the generative AI era.

Midjourney V8.2 becomes the default, prioritizing aesthetics and personalization

Midjourney announced that V8.2 has been released and is now the default model. The update focuses on aesthetics, personalization, and image quality, with a more creative, bolder, and “fresh” style, according to the company.

This is an important signal for creative and social media teams: image-generation tools are moving quickly from “good enough” to “distinctive.” Brands that make good use of prompts, style guides, and internal review systems will have a clear advantage in fast image production.

Claude Opus 5 appears on Perplexity

Perplexity said Claude Opus 5 is now available on Perplexity and Perplexity Computer. According to the platform, the model was compared with six other models on WANDR and ranked behind Fable 5 but at 57% lower cost.

What stands out is that intermediary platforms like Perplexity are becoming places where users experience models in a work context rather than just “trying chat.” This opens opportunities for marketers to optimize research, monitor competitors, and summarize trends faster.

New benchmark puts Claude Opus 5 at the top in knowledge work

Artificial Analysis said Claude Opus 5 leads the AA-Briefcase benchmark for agentic knowledge work, outperforming Claude Fable 5 by a significant Elo margin while reducing cost per task. The evaluation highlights large-document processing, report generation, presentations, and spreadsheets.

For Vietnamese businesses, this is the kind of capability with the most direct impact: report synthesis, market research, proposal writing, sales data analysis, and planning support.

Anthropic emphasizes security safeguards for Opus 5

Some cybersecurity community posts said Opus 5 was announced at a price of $5 per one million input tokens and comes with safety measures designed to allow source-code vulnerability discovery while preventing exploitation, pentesting, or binary scanning behavior.

This reflects a new trend: the more powerful AI becomes, the more sophisticated its safety barriers must be. For businesses, model selection is no longer just about quality, but also about risk control, especially with sensitive data.

OpenAI Developers said ChatGPT web users can create a share link for their “custom pet” so friends can adopt it. This is a feature with a strong community and personalization angle.

ChatGPT web lets users share customizable “pets”

From a product strategy perspective, features like this make AI feel less rigid and increase user attachment. For marketers, it suggests a way to design brand experiences: personalized, shareable, and with a sense of “mine.”

Palantir continues telling the story of unified data and “digital twins”

Content shared around Palantir emphasizes that if AI is to produce useful decisions, it must connect fragmented data systems into a living picture. The story of ontology, digital twins, and an orchestration layer built on top of legacy systems is central to the company’s positioning.

This is a highly relevant perspective for large enterprises in Vietnam, where marketing, sales, CRM, and operations data are often still scattered. AI only becomes truly effective when the data is clean enough, connected enough, and up to date.

Claude Opus 5 ranks No. 1 on SWE-Bench, according to Cline

Cline said Claude Opus 5 reached No. 1 on SWE-Bench with 97% and matched the Fable 5 group at a lower price. SWE-Bench is closely watched by the programming community because it reflects real-world software problem-solving ability.

For marketers using AI for automation, web scraping, tooling, or building internal chatbots, better coding ability means a significant boost in campaign testing speed, landing page creation, and data pipelines.

Grok speeds up its release cycle with 4.6 and 4.7 milestones

According to a post circulating on X, Elon Musk said Grok 4.6 will arrive in 2 weeks and Grok 4.7 in 4 weeks. The community sees this as a sign that xAI is accelerating its update cadence to stay competitive among frontier models.

Shorter release cycles are now a familiar signal in the AI industry. For businesses, relying on a single model is becoming riskier; it is better to build a multi-vendor strategy so you can switch flexibly as quality, price, and features change.

The SpaceX – Tesla – xAI scenario continues to fuel community discussion

Some X accounts spread a theory about a unified ecosystem between Tesla, SpaceX, and xAI, emphasizing the power of data, satellites, self-driving cars, robots, and energy. Although this is only community speculation, it shows that the market still sees Elon Musk’s vertically integrated model as a compelling story.

For marketers, the lesson here is the importance of an “ecosystem,” not just individual products. AI companies that win big are often the ones that connect data, distribution, and user experience in one closed loop.

Grok Odyssey is expected to launch this year

The community continues to discuss pressure around Grok Odyssey, a version that is said to appear by the end of the year. While there is no official information from a source in this data, the repeated appearance of the product name shows market expectations for xAI’s next leap.

What marketers should track is not isolated rumors, but the trend: users increasingly expect models not only to answer well, but also to work well in real-world scenarios, from search to computer operation.

Anthropic is seen as stronger in many real-world task tests

Some community reviews say Opus 5 performs well in terminal coding, novel problem solving, computer use, and business workflows. However, it is still seen as weaker in some areas such as legal, medical, or certain specialized benchmarks.

The main message here is that no model can dominate every task. Businesses and marketers should choose models by use case: writing, summarization, analysis, coding, or self-running agents all require different evaluation criteria.

Opus 5 is said to outperform the standard on ARC-AGI-3

Greg Kamradt said Opus 5 has reached a new leading position on ARC-AGI-3, surpassing the previous mark and showing a higher level of autonomy in logic tests. Even so, the post also acknowledges that the model still makes mistakes and has not fully solved the hardest demos.

For performance marketers, this underscores an important reality: AI can be very strong in many intermediate steps, but human oversight is still needed at the final verification stage.

The open source story continues to be linked to Elon Musk’s image

A post repeated that Elon Musk has long supported an open spirit, from naming OpenAI to opening technology, publishing recommendation algorithms, and saying he would open-source the X platform. While this framing is colored by personal support, it reflects the long-term communications value of “transparency” in technology.

For AI brands, open source is not just a technical choice but also a trust strategy. The market, especially large enterprises, often trusts platforms that are easier to verify and integrate.

Anthropic is putting direct pressure on premium model pricing

Many market accounts describe Opus 5 as a “Fable 5-style” model at less than half the price. Although the comparison is marketing-driven, the overall message is clear: the premium AI race is shifting toward optimizing value per task.

This is an opportunity for Vietnamese marketing teams, which are highly cost-sensitive. When quality is high enough and costs fall, it becomes easier to scale AI in content production, research, sales support, and customer service.

Opus 5 is described as a major step forward in agentic coding

Some comments say Opus 5 outperforms Fable 5 in agentic coding tasks while also offering a cost advantage. This view highlights a very important point: coding agents are becoming the central battleground for AI labs.

For digital marketers, strong coding agents can significantly shorten the time needed to build landing pages, automate reporting, integrate APIs, and create internal tools without a large engineering team.

Anthropic launches Opus 5 with a near-top but cheaper message

Many AI community responses said Opus 5 scores around 30 on ARC-AGI-3 and shows major progress in computer use. The core message is that the model can work with computers, not just answer inside a chat window.

This is a trend marketers should watch closely, because AI that can “operate” will unlock a new layer of automation: filling forms, looking up information, cross-checking data, and supporting multi-platform campaign operations.

Claude Opus 5 officially joins Anthropic’s new model lineup

Many market accounts confirmed that Anthropic has released its latest model, Opus 5. Although the post is brief, it remains a notable milestone because it shows Anthropic’s product line is being refreshed quickly and continuously.

The pace of launches like this forces professional users to stay updated on benchmarks, pricing, and integration capabilities rather than relying only on the reputation of an older model.

Opus 5 is described as near frontier but with significantly lower per-task cost

Some posts emphasize that Opus 5 delivers near-frontier performance, lower task-based cost, and is the default model for Claude Max. Community tests also note its self-checking and visual task handling abilities, such as reconstructing an object into a 3D model when the image cannot be viewed directly.

This is the kind of capability that fits modern marketing needs well: not only writing content, but also reading images, classifying assets, supporting creative ops, and checking consistency across multiple formats.

The open source debate continues to fuel speculation

A joking post asked what would happen if Anthropic accidentally released Opus 5 as open source. Although it is only a meme-style story, it reflects a real reality: pressure for transparency and verifiability is greater than ever.

For marketers, this means market trust will increasingly be tied to how explainable, controllable, and integrable the technology is.

Anthropic positions Opus 5 as close to Fable 5 but cheaper

Anthropic’s launch announcement emphasizes that Opus 5 delivers intelligence close to Fable 5 at only half the price. In a market where performance is gradually converging, pricing will be the decisive competitive lever.

This is the moment for Vietnamese businesses to reassess the ROI of AI: with the same budget, the model that delivers the greatest measurable impact will win.

OpenAI Developers said the ChatGPT Work agent can now use websites that require login. Users can switch to controlling a cloud browser to log in once, then let the agent continue working and keep the login state across sessions.

ChatGPT Work agent supports websites that require login

This is a very practical change. It moves AI from a “suggestion assistant” to an “action assistant,” well suited for repetitive marketing tasks such as checking dashboards, updating systems, and operating SaaS platforms.

Cursor integrates Claude Opus 5 for coding, with zero data retention support

Cursor said Claude Opus 5 is now available on its platform, scoring nearly the same as Fable 5 on CursorBench but at half the price and compatible with zero data retention. This is a major plus for enterprise users concerned about security.

For marketing teams with digital products, landing pages, or internal tools, the integration between AI and a secure coding environment can shorten experimentation time while keeping data under control.

Midjourney acquires Co-Star, expanding its product design ambitions

Midjourney said it is acquiring Co-Star and inviting CEO Banu Guler to serve as Chief Design Officer. At the same time, the company will build a cross-functional team covering design, product, and frontend, while the Co-Star app will continue operating under separate control.

This move shows that AI companies increasingly see experience design as a strategic advantage, not just a presentation layer. For marketers, it is a reminder that even the best tool needs a strong experience to gain broad user adoption.

What this means for the Vietnamese market

The news flow above shows AI shifting from a race over “who is smarter” to “who delivers better real-world efficiency at lower cost.” This is a major opportunity for Vietnamese businesses, especially in marketing, e-commerce, customer service, and content operations.

In the short term, marketers should focus on four things: first, standardize AI workflows for research, copy, visuals, and reporting; second, test models by task rather than by brand name; third, review data policies when integrating agents into systems; and fourth, invest in prompt, review, and measurement capabilities to avoid relying on intuition alone.

The most important thing is not to wait for a “super model” to solve everything. Competitive advantage will belong to teams that know how to combine the right model, clean data, clear processes, and people who are capable of controlling the output.

See more marketing news and guides at https://marketing365.vn.

Follow more updates from Marketing365 to stay up to date with the latest marketing trends.

Read more articles in the same category AI developments.

This article focuses on the latest AI news with a perspective for the Vietnamese market.