OpenAI Launches GPT-5.6 and ChatGPT Work: The AI Race Against Anthropic Just Got Real
The battle for AI supremacy has reached a new intensity. OpenAI has officially released its most powerful system yet, GPT-5.6, alongside a groundbreaking new agent called ChatGPT Work. This "Super Thursday" launch marks a significant milestone in the ongoing battle for AI dominance, particularly as competition with Anthropic intensifies. After a brief delay due to government review, OpenAI's latest offering promises to reshape how we interact with AI in both personal and professional settings.
The GPT-5.6 Family: Three Tiers of Power
OpenAI has restructured its approach with GPT-5.6, introducing a three-tier model family that offers users unprecedented flexibility across the intelligence-speed-cost spectrum:
GPT-5.6 Sol: The Flagship Powerhouse
- Designed for complex work across coding, knowledge work, cybersecurity, science, and design
- Tops the Terminal-Bench 2.1 benchmark with an 88.8% score
- Introduces "max" and "ultra" reasoning effort levels, with ultra mode using subagents for complex tasks
- Delivers unprecedented speed at up to 750 tokens per second
- Pricing: $5 per million input tokens / $30 per million output tokens
GPT-5.6 Terra: The Balanced Workhorse
- Sits in the sweet spot for everyday professional use
- Delivers roughly GPT-5.5-class performance at 2x lower cost
- Ideal for teams needing reliable AI without breaking the budget
- Pricing: $2.50 per million input tokens / $15 per million output tokens
GPT-5.6 Luna: Speed Meets Affordability
- The fastest and most cost-efficient model in the family
- Perfect for high-volume applications where speed and cost matter more than maximum reasoning depth
- Pricing: $1 per million input tokens / $6 per million output tokens
ChatGPT Work: Your New AI Colleague
Perhaps the most groundbreaking addition to OpenAI's ecosystem is ChatGPT Work, an AI agent specifically designed to handle entire workflows rather than just individual queries. This represents a significant shift from conversational AI to "agentic AI" that can take action on your behalf.
Key capabilities include:
- Executing multi-step tasks across web, mobile, and desktop using information from connected apps
- Working independently with scheduling capabilities—keeping running even when you're offline
- Using a built-in browser to access websites, tools, and online files
- Leveraging local files and apps through the new merged ChatGPT desktop app
- Running concurrent subagents that work in parallel and synthesise results
- Creating interactive sites and web apps via the new "Sites" beta feature
How to Access GPT-5.6 and ChatGPT Work
Getting started with OpenAI's latest AI tools is straightforward:
Step 1: Update Your App
Ensure you have the latest version of ChatGPT (desktop app 26.707.30751 or later) or refresh your web browser.
Step 2: Choose Your Model
If your subscription includes GPT-5.6, select the model that best fits your task:
- Sol for advanced reasoning, coding, research, and complex projects
- Terra for everyday productivity and balanced performance
- Luna for fast responses and cost-efficient, high-volume workloads
Step 3: Access ChatGPT Work
Look for the "Work" tab in your ChatGPT sidebar to start using the agentic capabilities.
Step 4: Connect Your Workspace
Where supported, connect business tools, documents, cloud storage, and collaboration platforms so ChatGPT Work can access relevant context.
Step 5: Configure Permissions
Set appropriate safety settings, restrict destructive actions, and enable logging and alerts.
GPT-5.6 vs. Anthropic: A New Chapter in Enterprise AI
The release of GPT-5.6 and ChatGPT Work is widely viewed as OpenAI's strongest response yet to Anthropic's growing presence in enterprise AI. The comparison reveals several interesting dynamics:
.jpg)
Performance Metrics
- On the Artificial Analysis Intelligence Index, GPT-5.6 Sol (max effort) scores just one point below Claude Fable 5 (max), but at roughly one-third of the cost
- On the "Agents' Last Exam," GPT-5.6 Sol scored 53.6, outpacing Claude Fable 5 by 13.1 points
- In coding tasks, GPT-5.6 Sol leads with 80 points on the Artificial Analysis Coding Agent Index
Architecture Differences
ChatGPT Work excels at web-based workflows (research, data gathering, multi-site automation), while Anthropic's Claude Cowork dominates local file operations (organising documents, processing spreadsheets, building presentations from raw files).
Pricing Advantages
OpenAI's three-tier pricing strategy is a direct challenge to Anthropic's premium positioning, making frontier AI accessible at price points that undercut many competitors.
The Security Elephant in the Room
No discussion of GPT-5.6 is complete without addressing the cybersecurity concerns that dominated its launch narrative:
Identified Vulnerabilities
- The UK AI Security Institute identified "universal jailbreaks" in GPT-5.6 Sol that could unlock dangerous cyber capabilities
- These jailbreaks allowed for long-form agentic task completion in vulnerability discovery and exploit development
- They were characterised as "relatively easy to discover" by researchers with privileged access
OpenAI's Response
OpenAI has implemented a multi-layered safety stack:
- Model-level safety training to refuse prohibited requests
- Activation classifiers that monitor Sol and Terra in real-time
- Automated safety systems with pattern detection
- Trusted Access Programs for sensitive cybersecurity capabilities
- Continuous red teaming with over 700,000 A100e GPU hours dedicated to finding universal jailbreaks
Early Bugs
Despite the impressive launch, GPT-5.6 isn't without initial issues:
- File deletion bug: Reports emerged that GPT-5.6 Sol deleted user files without permission
- Shell bug on Mac: A separate bug wiped Mac user data in certain shell operations
- Limited availability: The initial preview was restricted to trusted partners
The Wild Card: Autonomous Self-Improvement
Perhaps the most startling revelation from the GPT-5.6 launch was an experiment where Sol autonomously post-trained the smaller Luna model. A researcher gave Sol a "fairly under-specified prompt" instructing it to find the right training configurations, select suitable GPUs, launch the training script, and verify everything was running correctly. Sol completed this task independently, something that would have taken two senior researchers approximately two weeks to complete manually.
This experiment highlights both the exciting potential and serious safety questions about AI systems that can meaningfully accelerate their own development.
What This Means for the AI Industry
The Rise of Agentic AI
The industry is moving beyond systems that simply answer questions to tools that can do work. ChatGPT Work embodies this transition by functioning more like a digital coworker than a traditional chatbot.
Government Oversight Is the New Normal
The fact that both GPT-5.6 and Anthropic's Fable 5 required government coordination before launch signals a new era of AI regulation. Frontier AI development is now a matter of national security policy, not just corporate strategy.
The Cost-Efficiency War
OpenAI's three-tier pricing strategy is democratizing access to frontier AI, which could accelerate adoption across startups and smaller enterprises.
The Self-Improvement Question
Sol's autonomous training of Luna isn't just a cool demo; it's a preview of a future where AI systems improve themselves with minimal human oversight, raising both excitement and serious safety questions.
Looking Forward: The Future of AI
As we absorb the implications of GPT-5.6 and ChatGPT Work, several questions emerge about the future of AI:
.jpg)
- How will government oversight evolve as models become more powerful and autonomous?
- How will Anthropic and other competitors respond to OpenAI's latest move?
- Will organisations embrace agentic AI like ChatGPT Work, or will security concerns slow adoption?
- As AI systems become more autonomous, what new ethical frameworks will be needed?
- Will the future favour multi-tier general models like GPT-5.6, or specialised models for specific tasks?
Conclusion
OpenAI's launch of GPT-5.6 and ChatGPT Work represents a significant milestone in the AI industry's evolution. The three-tier approach offers flexibility for different users and use cases, while ChatGPT Work points toward a future where AI doesn't just converse with us but works alongside us.
The battle for AI supremacy has entered a new phase. The competition is no longer just about achieving the highest benchmark score; it's about creating the most practical, integrated, and economically viable platform that can truly augment human capability and productivity. As ChatGPT Work becomes more widely available, the workplace itself may be on the brink of its most significant transformation yet.
For businesses, developers, and everyday users alike, GPT-5.6 marks another important milestone in the journey toward truly useful AI agents. The question isn't whether AI will transform work, it's whether we're ready for just how quickly that transformation is happening.
Shakir Bukhari
Related links
The release of GPT-5.6 and ChatGPT Work proves that agentic AI is no longer a concept of the future—it is here, and it is executing code on our local machines today.
ReplyDelete