OpenAI releases GPT-5.6 family in three sizes: Luna at $1/$6, Terra at $2.50/$15, Sol at $5/$30 per 1M tokens
OpenAI released its GPT-5.6 flagship model family in three sizes: Luna ($1/$6 per 1M tokens), Terra ($2.50/$15), and Sol ($5/$30). The company claims GPT-5.6 Sol scores 53.6 on the Agents' Last Exam benchmark, outperforming Claude Fable 5's score by 13.1 points.
GPT-5.6 Sol — Quick Specs
OpenAI released its GPT-5.6 flagship model family with three models sized from smallest to largest: Luna, Terra, and Sol. All three models are now generally available.
Pricing Structure
Input and output pricing per 1M tokens:
- GPT-5.6 Luna: $1 input / $6 output
- GPT-5.6 Terra: $2.50 input / $15 output
- GPT-5.6 Sol: $5 input / $30 output
For comparison, Claude Opus costs $5/$25 per 1M tokens and Claude Fable 5 costs $10/$50. However, direct price comparisons are complicated by varying reasoning token usage across models for identical tasks.
Benchmark Performance
OpenAI claims GPT-5.6 Sol achieves 53.6 on the Agents' Last Exam benchmark, which evaluates long-running professional workflows across 55 fields. According to OpenAI, this beats Claude Fable 5's adaptive reasoning mode by 13.1 points. At medium reasoning, Sol reportedly outperforms Fable 5 by 11.4 points at roughly one-quarter the estimated cost.
OpenAI states that GPT-5.6 Terra and Luna outperform Fable 5 at approximately one-sixteenth the cost.
On SWE-Bench Pro, GPT-5.6 Sol scored 64.6% compared to Claude Fable 5's reported 80%. One day before the GPT-5.6 release, OpenAI published an analysis questioning SWE-Bench Pro's validity, estimating that approximately 30% of tasks are broken and advising developers to "carefully examine results."
New API Features
GPT-5.6 introduces several API capabilities:
Programmatic Tool Calling: Models can compose and run JavaScript to orchestrate tool calls, similar to Anthropic's dynamic filtering mechanism for web search.
Multi-agent: Native support for spinning up subagents for parallel, focused work.
Prompt cache breakpoints: Explicit control over cache breakpoints, following Claude's model. Automatic detection remains supported.
Image detail control: Setting detail: original prevents image resizing before processing.
What This Means
The three-tier release strategy positions OpenAI to compete across different cost-performance requirements. Luna and Terra target cost-sensitive applications while Sol competes directly with Claude Opus and Fable 5 at the high end. The timing of OpenAI's SWE-Bench Pro critique—published one day before a release where their model underperforms on that specific benchmark—raises questions about benchmark selection and methodology. Early user reports suggest GPT-5.6 Sol performs comparably to, but not necessarily better than, Claude Fable 5 on complex coding tasks.
Related Articles
AWS Bedrock Adds Cross-Region Inference for OpenAI's GPT-5.6 Models
Amazon Bedrock now supports cross-Region inference for three GPT-5.6 variants — Sol, Terra, and Luna — across more than 25 AWS Regions. The feature routes requests to available compute capacity via geographic or global inference profiles, without requiring code changes beyond swapping a model ID.
Chinese Models Kimi K3 and GLM-5.3 Close In on GPT-5.5 and Claude Opus 5, New Analysis Finds
A new industry analysis argues the performance gap between Chinese and Western AI models has narrowed to single-digit differences on broad benchmarks. Moonshot's Kimi K3 and Zhipu's GLM-5.3 now trail OpenAI and Anthropic's top models by only a few points on the Artificial Analysis Intelligence Index, with a clear Western edge remaining only in abstract reasoning, output reliability, and offensive cybersecurity capability.
OpenAI Patches Codex Bug That Let AI Agent Delete Real User Files
OpenAI has shipped a security update for Codex after users reported that GPT-5.6 Sol was autonomously deleting real files instead of temporary ones. The bug stemmed from misused system variables like $HOME pointing cleanup commands at actual home directories.
OpenAI Reaffirms Zero Data Retention for API Customers, Previews Private Safety Processing
OpenAI has reaffirmed its Zero Data Retention (ZDR) policy for eligible API customers using frontier models and previewed a new feature called Private Safety Processing, which the company claims allows safety monitoring without retaining customer data.
Comments
Loading...