OpenAI's GPT-6 SOL and Luna halve costs compared to GPT-5.6

OpenAI: OpenAI develops and deploys advanced AI systems including large language models. The company released the GPT-6 series featuring Sol and Luna variants that emphasize cost reductions. These models update the GPT lineup with changes across multiple evaluation benchmarks. GPT-6 Sol: GPT-6 Sol is a proprietary language model in OpenAI's GPT-6 family. It delivers halved pricing relative to GPT-5.6 while maintaining similar overall intelligence scores with gains in select coding tasks. The model also shows reduced hallucination rates on knowledge benchmarks. GPT-6 Luna: GPT-6 Luna is a lower-cost proprietary language model within the GPT-6 series from OpenAI. It offers substantial price cuts compared to its predecessor alongside mixed results across evaluations. The model reduces hallucination frequency while showing regressions in certain knowledge work metrics. SWE-Atlas-QnA: SWE-Atlas-QnA is a software engineering question-and-answer benchmark used in agentic coding evaluations. GPT-6 Sol shows gains on this test while GPT-6 Luna experiences a decline. Results contribute to the overall Coding Agent Index assessment. AA-Omniscience: AA-Omniscience is a knowledge and hallucination benchmark that penalizes incorrect answers and rewards abstention. Both GPT-6 models achieve lower hallucination rates on this evaluation than their predecessors. The benchmark tracks accuracy alongside answer attempt rates. Terminal-Bench 4.0: Terminal-Bench 4.0 is a specific evaluation focused on terminal command and coding agent performance. Both GPT-6 Sol and Luna record improvements on this benchmark relative to the previous generation. It forms one component of the broader Coding Agent Index. Artificial Analysis Coding Agent Index: The Artificial Analysis Coding Agent Index measures AI performance on end-to-end software engineering tasks. It highlights a modest gain for GPT-6 Sol and a small decline for GPT-6 Luna versus their GPT-5.6 counterparts. The index incorporates results from harnesses such as Codex. Artificial Analysis Intelligence Index: The Artificial Analysis Intelligence Index is an independent benchmark that aggregates multiple evaluations of AI model capabilities. It serves as the primary metric for comparing GPT-6 Sol and Luna against prior versions in this announcement. Scores on the index remained broadly level for the new models. Model Release: OpenAI launched the GPT-6 Sol and Luna models as updates to its frontier language model lineup. Efficiency Gains: The new models achieve meaningful reductions in cost per task across intelligence and coding evaluations. Benchmark Dynamics: Performance across evaluations shows a combination of targeted improvements and selected regressions.