OpenAI GPT-5.6 Sol, Terra, and Luna launch on Amazon Bedrock with 80-point Coding Agent Index score

OpenAI's GPT-5.6 model family is now generally available on Amazon Bedrock, introducing a three-tier naming system where the number identifies the generation and Sol, Terra, and Luna represent distinct capability tiers.

Performance claims and benchmark scores

GPT-5.6 Sol, the flagship reasoning model, claims to set state-of-the-art performance on multiple benchmarks. According to OpenAI, Sol scores 80 points on the Artificial Analysis Coding Agent Index, 2.8 points above the next-best model while using less than half the output tokens, taking less than half the time, and costing approximately one-third less.

On ExploitBench for cybersecurity research, Sol scores 73.5% versus 47.9% for GPT-5.5 at comparable output-token budgets. On Agents' Last Exam, an evaluation covering 55 professional workflow fields, Sol achieves 53.6 points, outperforming the next-best model by 13.1 points. At medium reasoning effort, it leads by 11.4 points at roughly one-quarter the estimated cost.

Sol introduces "max reasoning effort" settings for complex multi-step tasks including autonomous coding agents, vulnerability research, and drug discovery workflows.

The three-tier system

GPT-5.6 Terra serves as the balanced model for production workloads, delivering performance superior to GPT-5.5 at lower cost. OpenAI positions Terra for code generation, content workflows, structured data extraction, and general-purpose agentic tasks.

GPT-5.6 Luna targets high-volume inference with focus on speed and cost efficiency for classification, summarization, routing, and real-time applications where latency matters.

Pricing matches OpenAI's first-party rates and counts toward existing AWS commitments. Specific per-token pricing was not disclosed.

Infrastructure and availability

Amazon Bedrock's next-generation inference engine pools capacity to handle burst traffic patterns typical of agent workloads. The service introduces prompt caching with explicit cache breakpoints, offering a 90% discount on cached input tokens. Cached context remains available for at least 30 minutes.

In-Region inference keeps requests within specified AWS Regions for data residency compliance. The platform uses zero-operator access security enforced at the chip level, with all model calls governed by AWS IAM policies and logged in CloudTrail.

Classifier-flagged traffic data will be retained for up to 30 days for automated abuse detection, as required by the model provider.

Regional availability

GPT-5.6 Sol is available in US East (N. Virginia) and US East (Ohio). Terra and Luna are available in US East (N. Virginia), US East (Ohio), and US West (Oregon).

OpenAI simultaneously launched ChatGPT Work, a new agent for multi-step tasks, and updated ChatGPT desktop applications for Mac and Windows. Users can configure these applications to use GPT-5.6 through the Responses API on Amazon Bedrock.

What this means

OpenAI's shift to a tiered naming system with Sol, Terra, and Luna indicates a strategic move toward product segmentation based on capability and cost rather than sequential version numbers. The three tiers allow organizations to match model selection to specific workload requirements and budget constraints. The claimed performance gains on coding and security benchmarks position Sol as OpenAI's most capable model to date, though independent verification of these benchmarks remains pending. The AWS partnership expands OpenAI's enterprise reach through existing AWS relationships and commitment structures.