China’s DeepSeek has once again disrupted the artificial intelligence landscape. The Hangzhou-based company quietly released its DeepSeek V4 preview this week, bringing two new open-source models that challenge the dominance of OpenAI‘s ChatGPT, Google‘s Gemini, and Anthropic‘s Claude.

This latest DeepSeek V4 preview arrives as a direct competitor to the most advanced proprietary AI systems. The company has released two versions: V4-Pro (Expert mode) and V4-Flash (Instant mode). Both models share a massive one-million-token context window, allowing them to process entire books or extensive codebases in a single session.

DeepSeek V4 Pro Specifications and Performance

The V4-Pro model is a behemoth with 1.6 trillion total parameters, though it activates only 49 billion during inference. This efficiency allows it to rival top closed-source models while remaining accessible to developers. The smaller V4-Flash variant features 284 billion total parameters with 13 billion active, making it more practical for local deployment.

Both models are available on Hugging Face for download. However, running V4-Pro locally demands significant VRAM resources. The V4-Flash version offers a more realistic option for individual developers and smaller teams.

According to DeepSeek’s official announcement, the V4-Pro achieves a Codeforces rating of 3,206, surpassing GPT-5.4‘s 3,168 and Gemini 3.1’s 3,052. This positions it as the strongest open model for competitive programming tasks currently available.

How DeepSeek V4 Performs Against ChatGPT, Gemini, and Claude

Coding and Agentic Task Benchmarks

On LiveCodeBench, the V4-Pro scores 93.5 percent, outperforming Claude Opus 4.6’s 88.8 percent and Gemini’s 91.7 percent. For agentic tasks measured by Toolathlon, it achieves 51.8 percent, beating both Claude (47.2 percent) and Gemini (48.8 percent). The V4-Flash variant matches the Pro version on simpler agent tasks while consuming far less compute power.

However, the DeepSeek V4 preview does not lead in every category. Claude’s Opus 4.6 remains superior in long-context retrieval, scoring 92.9 percent on MRCR 1M compared to V4-Pro’s 83.5 percent. GPT-5.4 still tops Terminal Bench 2.0 with 75.1 percent accuracy versus V4-Pro’s 67.9 percent.

Mathematical Reasoning Capabilities

In mathematical reasoning, the results are mixed. V4-Pro achieves 95.2 percent on HMMT 2026 Math, slightly behind Claude’s 96.2 percent and GPT-5.4’s 97.7 percent. On IMOAnswerBench, it scores 89.8 percent, outperforming Claude (75.3 percent) and GPT-5.4 (91.4 percent) but trailing Gemini.

Cost Advantage: DeepSeek Disrupts AI Pricing

Where DeepSeek V4 preview truly changes the game is pricing. The V4-Pro costs just $3.48 per million output tokens. Compare this to OpenAI’s $30 and Anthropic’s $25 for equivalent workloads. That represents a cost reduction of roughly 85 to 90 percent.

This enormous gap makes DeepSeek extremely attractive for developers building AI-powered applications. For startups and enterprises alike, the savings could be transformative. The open-source nature of both models also eliminates vendor lock-in concerns.

Building on this pricing advantage, DeepSeek has positioned itself as the budget-friendly alternative to American AI giants. The company’s strategy mirrors its previous releases, which similarly undercut competitors on price while delivering competitive performance.

What This Means for the AI Industry

The arrival of the DeepSeek V4 preview signals a shift in the AI landscape. Open-source models are no longer just alternatives—they are direct competitors to proprietary systems. With performance matching or exceeding GPT-5.4 and Claude Opus 4.6 in key areas, DeepSeek proves that open development can rival closed ecosystems.

For developers, this means more choices and lower costs. The ability to download and run these models locally offers privacy advantages that cloud-based services cannot match. However, the hardware requirements for V4-Pro remain a barrier for many users.

Looking ahead, DeepSeek’s aggressive pricing and open-source approach will likely pressure competitors to reduce their own costs. The AI industry may see a price war similar to what happened in cloud computing over the past decade.

For more insights on AI model comparisons, check out our guide on the best AI models of 2026. You can also explore top open-source AI tools for developers and how AI pricing compares across providers.

Sony’s table tennis robot made me think about what happens when AI gets a body

I wanted to dismiss Sony’s table tennis robot as another expensive lab flex. A machine that can rally against elite players is impressive, sure, but it also sounds like the kind of demo built to make executives clap in a room where everyone already agreed to be impressed.

But table tennis is a nastier test than it looks. The ball is small, fast, spinning, and rude enough to change direction the moment it hits the table. Sony’s system faces something less forgiving than calculation. It has to see, predict, and act before the point is gone.

The challenge of embodied AI: why Sony’s robot matters

Sony tested Ace against five elite players and two professionals under official competition rules, and the robot came away with several wins. The more useful detail is what it had to handle during those matches: fast, high-spin shots that change direction after the bounce and punish even small delays. In plain English, Ace wasn’t just hitting the ball back. It was reading motion, making a prediction, and moving before the rally escaped it.

This is where the Sony table tennis robot transcends a simple sports demo. It becomes a case study in embodied AI — intelligence that must operate in the physical world, not just on a screen. Explore more AI robotics news.

AI is leaving the board

The usual “AI beats human” headline undersells what Ace is actually testing. We’ve already seen that story in cleaner arenas. IBM’s Deep Blue beat Garry Kasparov in 1997, and the symbolism still hangs over every old contest between human skill and machine calculation.

But chess, for all its strategic depth, is polite to computers. The board doesn’t wobble. The pieces don’t spin. A knight never comes screaming back at 60 miles per hour because someone clipped it at a nasty angle.

Sony’s robot points to a different shift. When AI has to move, intelligence becomes a timing problem. The system has to read the world quickly enough to act inside it. That’s more useful, and much harder to keep neatly boxed in.

How the body changes the problem for AI

This is where the table tennis demo starts doing more work. A robot that can track spin, predict motion, and adjust its response in real time isn’t automatically a factory worker, warehouse picker, nurse assistant, farmhand, or disaster-response machine. That leap would be too neat, which usually means it’s wrong.

The broader robotics market is already well past the cute-demo stage. The International Federation of Robotics says 542,000 industrial robots were installed in 2024, more than double the figure from a decade earlier. It expects installations to reach 575,000 in 2025 and pass 700,000 by 2028. That doesn’t make Ace a factory product, but it does make it part of a bigger automation story that’s already showing up on production floors.

On controlled industrial floors, robots need to handle variation instead of repeating one perfect motion forever. In logistics, they face crushed boxes, bad angles, missing labels, and people walking through the wrong lane at the worst possible time. Outdoors, mud, weather, uneven ground, and produce shaped by nature aren’t known for respecting software requirements.

The labor side of embodied AI

The labor side is where the story gets less cute. McKinsey estimates that today’s technology could theoretically automate activities accounting for about 57% of current US work hours. That isn’t a clean jobs-lost number, and McKinsey is careful about that point.

The pressure is subtler and probably messier: tasks get split apart, roles get redesigned, and some workers discover that “efficiency” has a habit of arriving with a spreadsheet and a forced smile. Read more about the future of work and automation.

Some settings raise the penalty for being wrong. A chatbot that gets something wrong can waste an afternoon. A robot that misreads a patient’s balance, a wheelchair, or a hospital hallway can do real damage. The more embodied AI becomes, the less forgiving its mistakes get.

The bill comes with the body: infrastructure costs

The infrastructure doesn’t disappear when AI gets legs, wheels, or a robot arm. It still depends on chips, data centers, cooling systems, electricity, water, and a grid that wasn’t built around every company suddenly discovering it needs more compute.

The International Energy Agency expects global data center electricity consumption to double to around 945 TWh by 2030, representing just under 3% of global electricity consumption. That share may sound small until a local grid, a water system, or a community near a new data center has to absorb the concentration.

It’s not all grim though. Smarter robots could reduce factory waste, help inspect dangerous sites, improve precision agriculture, and take on work that breaks human bodies for a living. The upside is real, but so is the cost.

Deep Blue made AI feel powerful inside a board game. Ace makes it feel like the board is gone, and the pieces are now factories, hospitals, farms, grids, and workers trying to guess what happens next.

Asimov imagined robots bound by rules. The version we’re actually building may be bound first by economics. Check out the latest robotics trends for 2025.

OpenAI GPT-5.5: ChatGPT takes a major step toward autonomous work

OpenAI has officially unveiled GPT-5.5, the latest iteration of its flagship AI model powering ChatGPT. This release marks a deliberate shift from simple conversational AI toward systems capable of handling complex, real-world tasks with minimal human guidance. The model is rolling out across ChatGPT and Codex for Plus, Pro, Business, and Enterprise users, with a premium “Pro” version reserved for higher-tier subscribers. As the company pushes toward autonomous work, GPT-5.5 signals a new era in how we interact with AI.

From answers to execution: the GPT-5.5 shift

Unlike earlier updates that focused on improving response quality, GPT-5.5 is engineered to handle multi-step tasks more effectively. It can interpret loosely structured prompts, plan workflows, execute actions, and self-check outputs—all with fewer iterations from the user. This means users no longer need to break down every request into tiny steps; the model does the heavy lifting.

OpenAI has positioned GPT-5.5 as a tool for AI productivity, not just conversation. It excels at coding, debugging, research, document creation, and data analysis across multiple tools and environments. In internal tests, the model completed complex workflows more efficiently, reducing the need for constant back-and-forth prompts. This is a clear move toward making ChatGPT a true enterprise AI workhorse.

Why GPT-5.5 matters for productivity

The release of GPT-5.5 underscores how rapidly AI development is accelerating. OpenAI only recently introduced GPT-5.4, yet it is already pushing forward with a system focused on real-world productivity. What makes GPT-5.5 noteworthy is not just its raw capability, but how it changes the user experience.

The model is designed to handle “messy” instructions—requests that are incomplete or loosely defined—and still produce structured outputs. This reduces friction for users who may not know how to craft precise prompts. For example, a developer could say, “Optimize this code for speed,” without specifying every variable, and GPT-5.5 would plan and execute the task.

OpenAI also claims significant improvements in reliability and safety, with stronger safeguards to reduce errors and boost output quality. These changes are crucial as AI tools become more embedded in professional workflows, where accuracy matters more than novelty. The launch comes amid increasing competition from companies like Anthropic, which are releasing advanced models focused on enterprise and security applications.

What GPT-5.5 means for users

Everyday users: a smoother experience

For everyday users, GPT-5.5 may feel like a smoother version of ChatGPT rather than a dramatic overhaul. The model requires less effort to use, as it can interpret broader instructions and deliver results without detailed prompts. This makes it more accessible for casual tasks like drafting emails, planning trips, or summarizing articles.

Developers and professionals: a new collaborator

For developers, researchers, and professionals, the impact could be more significant. GPT-5.5’s ability to plan, execute, and refine tasks makes it suitable for complex workflows, including coding projects, data-heavy analysis, and multi-step problem solving. Early use cases suggest that users are beginning to treat the model less like a search tool and more like a collaborator. Instead of asking one question at a time, they can assign a broader objective and let the system work through it.

This shift toward autonomous work is particularly valuable for businesses looking to scale operations. By reducing the need for constant human input, GPT-5.5 can help teams focus on strategic decisions while the AI handles routine tasks. For more on how AI is transforming business, check out our guide on AI productivity tools for enterprises.

What comes next for autonomous AI

GPT-5.5 is part of OpenAI’s larger push toward more autonomous AI systems. The company is increasingly focusing on models that can operate across tools, persist through longer tasks, and reduce the need for human intervention. Future updates are expected to expand these capabilities further, with deeper integrations into software ecosystems and improved ability to handle real-world workflows.

The long-term direction is clear: moving from reactive AI systems to proactive ones that can manage tasks with minimal input. As this shift continues, the key challenge will be balancing capability with reliability. GPT-5.5 shows that AI is becoming more capable of doing work, but its success will depend on how consistently it can deliver accurate and trustworthy results. For a deeper dive into AI trends, see our analysis on future AI trends in enterprise.

In summary, GPT-5.5 represents a critical step toward autonomous work in AI. It reduces user effort, improves efficiency, and opens new possibilities for professionals and businesses alike. As OpenAI continues to refine its models, the line between AI assistant and autonomous worker will only blur further. Learn more about ChatGPT enterprise features to see how this technology fits into your workflow.