GPT-5.6 Sol Preview: A Quiet But Major Shift Toward Agentic AI
The GPT-5.6 Sol preview is a controlled early release of OpenAI’s newest flagship agentic AI model, part of a wider GPT-5.6 model family that introduces Sol, Terra, and Luna with stronger autonomous coding, biology, and cybersecurity capabilities for select partners before broader public access. This is not just another incremental model bump; it is an intentional move toward AI systems that can own complex workflows rather than sit passively behind a prompt box. By limiting access to a small group of trusted partners through the API and Codex during the preview, OpenAI is signalling that agentic AI coding and advanced cyber and biology features are powerful enough to warrant a staged rollout. In practice, GPT-5.6 Sol looks less like a chatty assistant and more like an early software engineer and security analyst that happens to run on tokens.
Sol, Terra, Luna: A Model Family Built For Autonomous Workflows
OpenAI’s new model family splits GPT-5.6 into three tiers—Sol, Terra, and Luna—each tuned for different levels of capability and cost. Sol is the flagship, Terra is positioned as a balanced everyday model, and Luna is the faster, more affordable workhorse. OpenAI has also introduced a naming system where the number marks the generation and the names mark capability tiers that can evolve independently. That sounds cosmetic, but it matters: this is the architecture you build when you expect autonomous agents to be a product line, not a single monolithic model. Pricing reflects this stratification: Sol at USD 5 (approx. RM23) per million input tokens and USD 30 (approx. RM138) per million output tokens, Terra at USD 2.50 (approx. RM11.50) input and USD 15 (approx. RM69) output, and Luna at USD 1 (approx. RM4.60) input and USD 6 (approx. RM27.50) output. That quote-worthy spread telegraphs a clear intent: agentic capability will be priced as a premium tier.
| Model | Role in Family | Pricing (Input / Output) |
|---|---|---|
| Sol | Flagship, highest capability | USD 5 / USD 30 per 1M tokens (approx. RM23 / RM138) |
| Terra | Balanced, everyday work | USD 2.50 / USD 15 per 1M tokens (approx. RM11.50 / RM69) |
| Luna | Fast, lowest-cost option | USD 1 / USD 6 per 1M tokens (approx. RM4.60 / RM27.50) |
Agentic Coding: From Code Assistant To Autonomous Builder
The headline feature of GPT-5.6 Sol is agentic AI coding: the ability to plan, iterate, and coordinate tools in a way that starts to resemble an autonomous developer. OpenAI says Sol sets a new state of the art on Terminal-Bench 2.1, a benchmark focused on command-line workflows that demand planning, iteration, and tool coordination. Add the new "max" reasoning effort, which gives Sol more time to think through hard problems, and you get a model designed to own the entire coding loop from requirement to implementation to debugging. The "ultra" mode is even more aggressive: it uses subagents to work on complex tasks beyond a single-agent setup. In other words, OpenAI is now selling not just a single coder but a small autonomous dev team in a box. That is a direct challenge to how software engineering is structured inside enterprises.
Biology And Cybersecurity: High-Capability AI With High-Friction Guardrails
Sol’s biology and cybersecurity skills are where the preview feels less like a product demo and more like a policy experiment. OpenAI describes GPT-5.6 Sol as its most capable model yet for cybersecurity and notes improved performance across coding, biology, and cybersecurity. On GeneBench v1, Sol outperforms GPT-5.5 on long-horizon genomics and quantitative biology analyses while using fewer tokens. In security, the model shifts the performance–efficiency frontier for long-horizon tasks such as vulnerability research and exploitation, but OpenAI stresses that it is better at helping people find and fix vulnerabilities than carrying out end-to-end attacks. According to OpenAI, Sol, Terra, and Luna are all classified as High capability in Cybersecurity and Biological and Chemical risk under its Preparedness Framework, without crossing the Cyber Critical threshold or the high threshold for AI self-improvement. This is paired with layered safeguards—model-level refusals, real-time misuse classifiers, account-level review, differentiated access, monitoring, and continued testing. The message is blunt: you get serious AI cybersecurity features and biology analysis, but you will feel the friction.
Preview-Only Access Today, But Agentic AI Is Clearly The Future
For now, the GPT-5.6 Sol preview is confined to a small group of trusted partners with access through the API and Codex. Some of those early users will experience blocked requests or slower responses when generations are paused for extra review, especially in dual-use security contexts where defensive and offensive work can initially look similar. Behind the scenes, OpenAI reports more than 700,000 A100-equivalent GPU hours spent on automated red teaming aimed at universal jailbreaks, plus human and third-party tests. The company plans to keep testing during the preview and publish an updated system card when GPT-5.6 moves toward general availability, which it says will happen in the coming weeks, including broader access for ChatGPT, Codex, and API users. Ordinary users should expect two things: more powerful agentic behavior in coding and security tasks, and tighter safety checks that occasionally get in the way. But the direction of travel is obvious. GPT-5.6 Sol is not just a smarter chatbot—it is a step toward autonomous AI agents that will sit inside critical workflows by default.
