OpenAI Launches GPT-5.6 Models with Enhanced Capabilities
In brief
- OpenAI has released its new GPT-5.6 models-Luna, Terra, and Sol-available for widespread use starting today.
- These models are priced per 1 million input/output tokens, ranging from $1/$6 for Luna to $5/$30 for Sol.
- While the pricing is competitive with other leading models like Claude Opus, the real innovation lies in their capabilities.
- The GPT-5.6 series boasts a February 2026 knowledge cutoff and an impressive million-token context window.
- Its standout feature is its improved "long-running agentic performance," as demonstrated by beating Claude Fable 5 in tasks across 55 professional fields, including coding.
- However, while GPT-5.6 Sol excelled in these benchmarks, it underperformed compared to Fable 5 on the SWE-Bench Pro evaluation, where it scored 64.6% versus Fable 5's 80%.
- OpenAI has since flagged issues with SWE-Bench Pro, suggesting that nearly 30% of its tasks may be flawed.
- The models introduce new API features like Programmatic Tool Calling, which enables them to execute JavaScript and orchestrate tool calls, bridging the gap between model-based and terminal-based utility use.
- This could enhance their versatility for developers.
- As OpenAI continues to refine these tools, users can expect further improvements in efficiency and cost-effectiveness.
Terms in this brief
- Luna
- One of OpenAI's new GPT-5.6 models, designed for specific use cases with lower cost and fewer capabilities compared to more advanced models like Sol.
- Sol
- The most powerful model in OpenAI's GPT-5.6 series, known for its high performance across various professional tasks but also noted for higher costs.
- SWE-Bench Pro
- A benchmark test evaluating AI's ability to fix real bugs from open-source software projects on GitHub, considered a practical measure of coding skills.
Read full story at Simon Willison →
More briefs
AI Pharmacy Tool Causes Confusion
Kinney Drugs has started using an artificial intelligence tool to manage prescription refills. The tool has been calling and texting customers with confusing messages about their medications. The tool has caused many problems for customers. It has ordered the wrong medications and failed to notify customers when their refills are ready. This has led to delays and frustration for customers. About a dozen customers have reported problems with the tool. The use of this tool raises concerns about privacy and regulation. A new data privacy bill will take effect in two years, but it may not be enough to protect consumers. Customers will have to wait and see if the issues with the AI tool are resolved.
AI Breakthrough: Anthropic's Claude Opus 5 Creates Full 3D Games from Single Prompts
Anthropic has unveiled Claude Opus 5, an AI capable of generating complete 3D games directly from text prompts. Unlike previous systems, Opus 5 doesn't require external assets-it creates everything from geometry and textures to physics and even music entirely within the browser. Games like first-person shooters, kart racers, and Minecraft clones are generated in full detail, surpassing competitors GPT-5.6 Sol and Kimi K3 in complexity and depth. This advancement marks a significant leap forward for AI in game development, potentially reducing production time and costs. Developers could soon prototype games faster or create custom levels on the fly. Researchers see Opus 5 as a tool to explore new interactive storytelling methods and educational simulations. As AI capabilities continue to evolve, expect more innovations that blur the lines between creativity and technology. Future developments may include even more sophisticated interactions and personalized gaming experiences tailored to individual preferences.
AI-Backed Chess App Challenges Traditional Learning Methods
A new chess learning app, inspired by how large language models (LLMs) are trained, has launched. The app presents users with real game positions and two possible moves-one made by a human player and the other by Stockfish, a powerful chess AI. Users choose which move they believe is better, then receive immediate feedback on their decision along with an explanation of how Stockfish would play out the next few moves. This approach differs from traditional chess learning tools by focusing on imitation rather than rote memorization. While it may not be the most efficient way to learn the basic rules of chess, the app aims to quickly improve players' skills before they need more advanced practice, such as mastering openings or specific game strategies. The app also includes features like an Elo-style ranking system and options to share moves with friends or seek further explanations from AI tutors. The app's creator believes this method could be particularly effective for intermediate learners looking to refine their decision-making skills. While it’s unclear how it stacks up against other chess puzzle apps long-term, early users are already seeing improvements in their gameplay.
LanceDB Vector Database Boosts AI's Multimodal Smarts
LanceDB, a new vector database tailored for AI tasks, is making it easier for large language models to handle diverse data types like text, images, and audio. Unlike traditional systems that struggle with scattered information or mixed media, LanceDB natively supports multimodal data, streamlining how AI processes complex inputs. This advancement could unlock more efficient AI applications in fields like customer service, healthcare, and creative industries where diverse data is common. This innovation matters because it addresses a key limitation of current AI systems: their reliance on vector databases that aren’t optimized for handling multiple data types at once. By simplifying the integration of multimodal data, LanceDB could accelerate the development of more versatile AI applications. For developers, this means less time spent managing complex data pipelines and more focus on building innovative solutions. As LanceDB gains traction in the AI community, expect to see more creative uses of multimodal data across industries. Whether it’s enhancing customer interactions or improving diagnostic tools, LanceDB’s capabilities are poised to expand the possibilities for AI-driven advancements.
Irvine Expands Wildfire Detection System
Irvine is expanding its partnership with SensoRy AI to install a network of early wildfire detection sensors. The City of Irvine will start installation in early 2027. The new system can detect wildfires from over a mile away. This is important because it can help firefighters stop fires before they spread. The system was tested last year and worked well. The city is using funding from the California Department of Conservation to pay for the new system. The founder of SensoRy AI will continue to work on the project while attending Stanford University this fall. New wildfire detection sensors will be installed soon.