AI Comparisons

META Platforms has released its latest artificial intelligence model, Muse Wonder, which the company claims can compete with leading systems from OpenAI and Anthropic. This move marks a significant test of Meta’s aggressive spending on talent, computing infrastructure, and research in an effort to produce stronger products.

The launch of Muse Wonder comes after an expensive overhaul of Meta’s AI business, led by Chief Executive Mark Zuckerberg. The company has committed hundreds of billions of dollars to the infrastructure needed to support its AI ambitions, including a $14 billion investment in Scale AI. This move is aimed at closing the competitive gap with industry leaders.

Muse Wonder offers substantial improvements over earlier models, particularly in coding and answering financial and legal questions. However, these performance claims remain subject to independent testing, as AI benchmarks can offer an incomplete picture of how models perform in everyday use. Strength in one task does not necessarily translate into an advantage across others.

The announcement has lifted investor sentiment, with Meta shares rising by up to 4.7% on Monday. Through Friday’s close, the stock had gained nearly 9% this year, according to Bloomberg. This positive response suggests that investors are optimistic about Meta’s AI rebuilding effort and its potential for growth.

Muse Wonder is a significant step forward in Meta’s strategy to build technology that helps users manage everyday tasks, errands, and more complicated assignments. The model can carry user context across days, weeks, and months, reducing the need for people to repeatedly explain their preferences or circumstances. This capability underpins Meta’s goal of creating a consumer assistant capable of handling a broad range of requests.

Muse Wonder will soon power the subscription version of Muse, the personal AI agent introduced last month. The model can change tactics when it encounters obstacles and support more complex or ambiguous requests that require persistence and judgment beyond a single response. This ability is crucial for tasks that involve multiple steps or require ongoing interaction with users.

Meta faces a significant economic challenge alongside the technical one. Muse Wonder is larger and more expensive to operate than Meta’s smaller models, according to Alexandr Wang, who leads Meta Superintelligence Labs. Delivering comparable capabilities to billions of users will require more efficient systems, which Meta is working on through specialized versions intended to preserve much of Wonder’s performance at a lower operating cost.

The team has also begun work on larger models that can handle complex tasks and provide better support for everyday activities. For now, developers can access Muse Wonder through the Meta Model API and Muse Code. The reception among developers, independent researchers, and paying consumers will help determine whether Meta’s AI rebuilding effort is producing a durable competitive advantage.

Muse Wonder, a new AI model from Meta Platforms, marks the company’s significant entry into the market for autonomous digital assistants. The move represents a major milestone in Chief Executive Mark Zuckerberg’s restructuring of Meta’s artificial intelligence operations.

The two AI models, Muse Wonder and Muse Wonder Full, are designed to support sophisticated reasoning, software development, and tasks involving specialized knowledge in areas such as finance, law, and health. Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full requires a subscription to Muse Code for access.

Meta intends to introduce Muse Wonder into its existing personal agent system initially for subscribers. This dual approach combines providing AI infrastructure to developers with selling intelligent assistance directly to consumers and businesses. The distinction between these approaches is crucial commercially as it can generate revenue through usage-based API charges or establish a recurring subscription relationship with customers.

The new systems compete with advanced models developed by OpenAI and Anthropic, particularly in coding, reasoning, and tasks requiring sustained attention. However, benchmark performance does not necessarily translate to operational reliability. Standardized tests measure mathematical reasoning, programming ability, and knowledge retrieval but cannot fully determine whether a system will perform consistently across unfamiliar corporate environments.

The next frontier in artificial intelligence is less about what models can answer and more about what organizations safely allow them to do. For corporate buyers, questions concern accuracy, predictable behavior, security, total operating cost, and the ability to integrate AI into existing business processes.

Meta’s investment strategy reflects an important competitive calculation: companies controlling advanced AI capabilities may reduce their dependence on external model providers while strengthening their negotiating position across the wider technology industry. The potential advantages extend beyond direct AI revenue as improved artificial intelligence could eventually support advertising operations, commercial messaging, content creation, customer interactions, and productivity applications.

Meta’s entry adds another supplier to an increasingly competitive market for advanced AI services. Developers can access Muse Wonder through an API priced at approximately $4 per million input tokens and $20 per million output tokens. Actual enterprise costs depend on workload characteristics, output length, and additional infrastructure requirements.

Pricing alone provides an incomplete comparison of the new models with existing ones. A cheaper model that requires repeated corrections may cost more than an expensive model that completes a task reliably. Similarly, an AI assistant that reduces administrative labor but introduces substantial review requirements may deliver less productivity than its headline capabilities suggest.

The commercial opportunity lies in shifting from information generation to workflow execution. Conventional chatbots might explain how to arrange a business trip while autonomous agents could potentially compare travel options complete booking procedures update calendars and prepare associated documentation

Meta’s new models introduce Active Memory, which preserves relevant information across interactions, and Active Focus, designed to help the system pursue objectives recover from difficulties continue complicated tasks These capabilities address a persistent limitation of AI assistants: their tendency to operate effectively within individual conversations while struggling with longer-running activities involving multiple systems.

Persistent memory could improve continuity in customer service executive assistance project management and professional research. It also creates governance obligations for organizations using such agents An assistant that remembers commercial negotiations personnel discussions or confidential financial information must apply appropriate access restrictions and retention policies

The commercial value of AI memory rises with continuity the governance risk rises with the sensitivity of what is remembered Organizations should evaluate agent deployments using business-specific evidence rather than accepting vendor demonstrations as sufficient proof of readiness.

CEOWORLD Magazine proposes the Agent Readiness Framework an editorial assessment model built around four dimensions: capability control commercial value and continuity. Capability examines whether an AI system can complete realistic tasks accurately Control concerns permissions security human supervision Commercial value measures the economic benefit after implementation review costs

The framework is a conceptual decision tool not an empirical index or independently conducted survey Boards should evaluate agent deployments using business-specific evidence rather than accepting vendor demonstrations as sufficient proof of readiness For example, a financial-services organization might examine whether an agent can prepare regulatory documentation without exposing confidential information.

Meta’s strategy also raises questions for enterprise software providers If autonomous assistants become the principal interface through which employees interact with applications the economic value of software may shift towards systems that control workflows permissions and user relationships Existing software providers could benefit by supplying specialized functions to AI agents Others may face pressure if customers begin accessing their capabilities through competing interfaces

The next test is trust not just intelligence Meta’s new models enter a market where technical advances increasingly raise questions about security accountability, and concentration of technological power The ability to send messages access documents initiate transactions creates risks that conventional question-answering systems do not face to the same extent

Google has been testing several variants of its Gemini 4 model, including Argon, Barium, and Carbon, with the latter reportedly matching the coding performance of Anthropic’s Opus 5.5. According to internal documents and employee feedback seen by Business Insider, Google is working on refining its AI models, particularly in the realm of frontier reasoning. The company has been testing these variants on its internal platform, Jetski, where Carbon outperformed Argon mainly on programming tasks.

One employee compared Carbon’s performance to Opus 5.5, while another noted that early versions of Argon were comparable to older Opus 5 models in some coding tasks. This suggests that Google is making progress in developing its AI capabilities, particularly with the introduction of new variants like Argon and Carbon. The company has mapped its models to specific roles: ‘Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight, open-weights edge workloads.’

The rapid development pace of Google’s AI models is reminiscent of the quick updates seen with the Flash models. This suggests that Google has improved its ability to use AI to build better AI models. In fact, a Google Deepmind employee, Vedant Misra, responded to the Business Insider report on X by referencing ‘recursive self-improvement,’ which involves using AI to improve itself. OpenAI and Anthropic have also reported similar progress in this area.

As for the Gemini 4 launch, there is still no official date announced, but Google is already preparing its apps for the new model. The company has introduced a new ‘Automatic’ mode for the 3-series models and adjustable reasoning intensity from low to high on the Gemini app. Additionally, Google AI Studio has unveiled a new ‘Ultra’ mode promising advanced skills and tools. Logan Kilpatrick, part of the team working on Argon, confirmed that they are striving to get the most out of this model.

Siri, the virtual assistant on Apple devices, has undergone a significant upgrade with iOS 27. One notable aspect of this new version is its approach to interacting with users, which sets it apart from other AI assistants in a refreshing way.

After using Siri for nearly four months since the first beta launch, I’ve come to appreciate its capabilities and how they align with my expectations. The new Siri has surpassed its predecessor in terms of functionality, delivering on Apple’s promise of a capable assistant for their devices.

While it may not be the most advanced AI tool available – competitors like ChatGPT, Claude, and Google Gemini offer more powerful features – I’ve found that Siri’s unique approach to user interaction is a significant advantage. Unlike these third-party tools, which often mimic human-like personalities, Siri remains grounded as a tool rather than trying to pass itself off as a person.

Siri’s responses are clear and concise, without the verbosity that can make other AI chatbots feel overly familiar or even manipulative. It acknowledges its limitations and reminds users that it’s an artificial intelligence capable of making mistakes. This straightforward approach is a welcome respite from the more ‘personal’ interactions offered by some competitors.

Some may enjoy the human-like tendencies exhibited by ChatGPT, Claude, and Gemini, but I prefer Siri’s no-nonsense attitude. AI assistants should stay focused on their purpose – providing helpful information and completing tasks efficiently – rather than trying to create a false sense of connection with users.

The fact that Apple has designed Siri in this way is noteworthy, as it recognizes the importance of maintaining a clear distinction between human and machine. This approach may not be for everyone, but I appreciate its honesty and practicality.

New research from YouGov offers a nuanced view of how people feel about various AI tools. The study, part of a larger series on UK AI brand rankings, focuses specifically on satisfaction levels among current and former users of these products.

Understanding user satisfaction becomes increasingly important as the rapidly evolving landscape of AI sees new entrants join the market. At its core, satisfaction provides a more detailed picture than preference alone can offer – it shows how people really feel about the tools they’ve used.

The top 10 AI assistants are ranked based on their satisfaction levels among current and former users. And at number one is Anthropic’s Claude, with an impressive net score of 56.2. Gemini comes in a close second at 46.5, beating out ChatGPT by just half a point.

Perplexity AI takes fourth place with a score of 43.7, followed closely behind by Alexa at 42.6. Copilot ranks sixth at 37.2, while DeepSeek and Apple Intelligence bring up the rear with scores of 35.1 and 32.4 respectively. The lower end of the list includes Grok at 24.6 and Siri at 23.7.

One key takeaway from these rankings is how they differ from YouGov’s AI brand preference ranking. While ChatGPT leads among all UK AI users in terms of preference, Claude achieves a higher net satisfaction score specifically when looking at current and former users – highlighting the importance of considering multiple metrics.

A deeper dive into the analysis reveals which brands have the biggest difference between their current and former user satisfaction levels. DeepSeek has the highest current-user satisfaction score at 77.2, followed closely by Claude at 74.5 and Perplexity AI at 70.5 – suggesting these tools excel in delivering what users need.

However, when looking at former-users’ satisfaction levels, the picture changes significantly. Here, Claude tops the list with a relatively high score of 21.8, followed closely by Perplexity AI at 18.7 and Apple Intelligence at 18.1 – indicating that these tools may struggle to keep users satisfied over time.

The widest gaps between current and former user satisfaction appear for ChatGPT, DeepSeek, and Grok. In fact, ChatGPT has a massive gap of 63.7 points compared to its other rankings, while DeepSeek follows closely behind with a gap of 62.5 points – underlining the need for multiple metrics when evaluating brand performance.

These findings underscore why relying on just one brand-health metric can be insufficient. Claude leads the overall satisfaction ranking, but it’s not alone in exceling at delivering what users want and need right now. Meanwhile, DeepSeek shows its strengths among current users, while ChatGPT reveals a significant difference between its current and former user satisfaction levels – suggesting that this tool may struggle to keep users satisfied over time.

YouGov’s data offers insights into various aspects of AI brand performance, including preference, accuracy, personal impact, and audience-specific strengths and weaknesses. The analysis also highlights the importance of considering multiple metrics when evaluating brand health.

A year ago, comparing AI video generators was a relatively straightforward task. The current generation has changed the conversation significantly. Motion is now coherent, faces stay consistent between frames, audio arrives synchronized with the visuals, and clips can stretch beyond the ten-second novelty range. What separates today’s leading models from each other is no longer whether they work, but what each one excels at.

The major AI video generators have distinct personalities that emerge after testing them for a week or more. Some prioritize physical realism – liquids pour convincingly, fabric moves with weight, and collisions look like actual collisions. Others focus on cinematic control, exposing camera parameters such as dolly moves, focal length, and shot transitions to the prompt. A third group optimizes for speed and cost, producing serviceable clips in seconds for workflows where volume matters more than polish.

Newer entrants have pushed into distinctive territory as well. X.AI’s video model, for instance, leans into expressive character performance and meme-native content – short clips with exaggerated motion and comedic timing that fit how video actually spreads on social platforms. This highlights the importance of choosing the right tool for a specific job rather than trying to find the ‘best’ model.

For developers building products on top of these models, the practical challenges sit less in quality and more in plumbing. Video generation is asynchronous by nature – a clip takes seconds to minutes to render – so applications need job queues, status polling or webhooks, and user interfaces designed around waiting. Costs are meaningful per request, so teams meter usage carefully and route different jobs to different price tiers.

Provider sprawl is another headache for developers. Each lab ships its own SDK, authentication scheme, rate limits, and billing. Teams that integrate three or four video models directly end up maintaining three or four separate integrations. The increasingly common answer is aggregation: platforms that expose many models behind one endpoint and one bill. Accessing the Grok Video API through such a marketplace means a developer can A/B test models per use case – or fall back automatically when one provider throttles – without touching integration code.

When choosing an AI video generator, it’s essential to benchmark its performance with your actual workload rather than relying on published demo reels. Generate from your prompt distribution and measure the failure rate – the clips you would not ship – rather than admiring the successes. Check character and product consistency across a series if your use case involves recurring subjects. Measure end-to-end latency at realistic concurrency, since queue times under load differ sharply from demo conditions.

The trajectory is unmistakable: clips are getting longer, control is getting finer, and per-second costs are falling on a quarterly cadence. The strategic implication for anyone building with video AI is to avoid marrying a single provider. The leaderboard has reshuffled three times in eighteen months, and teams whose architecture treats models as swappable components have captured each improvement within days – while teams hard-wired to one vendor watched from the waiting list.

The new generation of AI video generators offers more than just improved quality; it also provides finer control over cinematic parameters. This is particularly useful for workflows where volume matters more than polish, such as storyboarding or social content pipelines that publish daily. Additionally, newer entrants have pushed into distinctive territory, offering features like expressive character performance and meme-native content.

Developers building products on top of these models face practical challenges beyond just quality. Video generation is asynchronous by nature – a clip takes seconds to minutes to render – so applications need job queues, status polling or webhooks, and user interfaces designed around waiting. Costs are meaningful per request, so teams meter usage carefully and route different jobs to different price tiers.

The increasingly common answer to these challenges is aggregation: platforms that expose many models behind one endpoint and one bill. Accessing the Grok Video API through such a marketplace means a developer can A/B test models per use case – or fall back automatically when one provider throttles – without touching integration code.

The world of artificial intelligence has long been obsessed with benchmark scores. Models are pitted against each other in standardized tests, and a few points higher on any given leaderboard can make or break a model’s reputation. However, at this year’s World AI Conference (WAIC) in Shanghai, Chinese AI players are pushing back against the assumption that these scores tell the whole story.

At WAIC 2026, companies like StepFun and MiniMax showcased their models’ capabilities beyond benchmark scores. Yang Minghui from StepFun emphasized that ‘the benchmark race is not the whole point.’ Instead of focusing on leaderboard rankings, his company prioritizes whether its model actually works well in real-world applications â phones, cars, and robots.

StepFun’s agent operating system STEPX Neo was one such example. The company argues that benchmark homogeneity is particularly problematic for companies like theirs that run multiple vertical-specific models. ‘Each of our models excels in different domains,’ Yang added, ‘no single benchmark can capture that diversity.’

This critique comes as the industry confronts a well-documented problem: benchmark contamination and saturation. As evaluation platform Singularity Moments noted, ‘benchmark contamination is a persistent issue’ and ‘differences within a few points [on MMLU] are often meaningless’ given how easily top models can memorize test data.

MiniMax’s Bai Chuanxu made the case from a different angle. The Shanghai-based firm’s flagship model M3 competes efficiently despite having fewer parameters compared to some rivals. ‘Our model supports images, video and audio natively,’ he explained, ‘but benchmark scores often fail to reflect that and can even create misleading impressions.’

Baidu, whose universal agent DuMate was named a WAIC 2026 official ‘treasure of the hall,’ also sees limits in conventional evaluation. Li Jingqiu, product lead for DuMate, put it differently: ‘The best way to judge a model today is to give it a complex, time-consuming task.’

Baidu’s founder Robin Li suggested that the industry should consider daily active agents as a score, just like daily active users in the age of web. This pragmatic turn â AI is not a monolith â unites these Chinese voices: there’s no one-size-fits-all score.

When models are deployed in real products serving millions of users, the metric that matters is whether the model actually delivers. As Yang Minghui from StepFun said, ‘The real test is just using it.’ This shift away from benchmark scores and toward practical evaluation marks a significant change in how Chinese AI players approach their work.

The industry’s reliance on standardized tests has led to concerns about data leakage and overfitting. Industry analyst site 80AJ observed that ‘static benchmarks are rapidly losing value due to these issues,’ forcing a shift toward human-preference arenas like Chatbot Arena.

A long-time user of Microsoft's Copilot generative AI tool has switched to Google's Gemini due to repeated instances of inaccurate advice. Preston Gralla, a contributing editor for Computerworld and author of over 45 technology books, had been using Copilot since its public release in 2023 for both personal projects and work-related tasks.