# OpenAI 预览 GPT-5.6 系列：Sol、Terra、Luna 三层模型及新推理模式 - 来源：MarkTechPost（RSS） - 作者：Michal Sutter - 发布时间：2026-06-27 03:18 - AIHOT 分数：77 - AIHOT 标记：同事件 - AIHOT 链接：https://aihot.virxact.com/items/cmqvbzwq80btusl80qqz9siu0 - 原文链接：https://www.marktechpost.com/2026/06/26/openai-previews-gpt-5-6-with-sol-terra-and-luna-tiered-models-new-reasoning-modes-limited-access ## AI 摘要 OpenAI 开始有限预览 GPT-5.6 系列，分为旗舰 Sol、日常生产 Terra 和低成本快速 Luna 三个层级。新增两种推理控制：max（加深单链推理）和 ultra（利用子智能体并行处理复杂任务）。在 Terminal-Bench 2.1 上，Sol (ultra) 得分 91.91%，Sol (max) 88.76%，超过 Claude Mythos 5（88%）和 GPT-5.5（83.4%）。定价方面，Sol 输入/输出每百万 token 为 $5/$30，Terra 为 $2.50/$15，Luna 为 $1/$6。目前仅向约 20 个合作伙伴开放 API 和 Codex 预览，更广泛接入计划在未来几周内推出。OpenAI 还计划在 7 月用 Cerebras 硬件使 Sol 达到每秒 750 token。 ## 正文 OpenAI has begun a limited preview of GPT-5.6, its next-generation model series. The lineup splits into three named tiers: Sol, Terra, and Luna. Sol is the flagship. Terra targets everyday production work. Luna is the fast, low-cost option. OpenAI is starting with a small group of trusted partners through the API and Codex. According to OpenAI post, they shared the models and plans with the U.S. government first. Broader access in ChatGPT, Codex, and the API is planned in the coming weeks. The change is mostly structural. GPT-5.6 introduces tiered models, two new reasoning modes, and a heavier safety stack. What is GPT-5.6? GPT-5.6 is a family, not a single model. OpenAI also changed how it names releases. The number now marks the generation. The names mark durable capability tiers. Each tier can advance on its own schedule. That gives developers a clearer choice across intelligence, speed, and cost. OpenAI calls Sol its strongest model yet. It cites gains in coding, biology, and cybersecurity. Terra matches GPT-5.5 performance while costing roughly half as much. Luna brings strong capability at OpenAI’s lowest price. New Reasoning Modes: max and ultra GPT-5.6 adds two reasoning controls. The first is a new max reasoning effort. It gives Sol the most time to reason deeply. The second is ultra mode. Instead of one model working alone, ultra leverages subagents. These subagents split complex work to accelerate it. Think of it this way. The max setting deepens a single chain of reasoning. The ultra mode coordinates several workers on one task. Both trade latency and cost for accuracy on long-horizon problems. Interactive Explainer Benchmark OpenAI shared a preview set of evaluations. Sol sets a new state of the art on Terminal-Bench 2.1. The benchmark tests command-line workflows that need planning, iteration, and tool coordination. 媒体内容 · 前往原文查看 Model / modeTerminal-Bench 2.1 GPT-5.6 Sol (ultra)91.91% GPT-5.6 Sol (max)88.76% Claude Mythos 588% GPT-5.583.4% source: venturebeat On Agent’s Last Exam, Sol was the only model past the halfway mark. It reached 50.9% in ‘code mode,’. On GeneBench v1, Sol beat GPT-5.5 on long-horizon genomics analysis. It did so while using fewer tokens. On ExploitBench, OpenAI reports Sol was competitive with Mythos Preview using about one-third of the output tokens. Pricing and Access GPT-5.6 is priced per one million tokens. Caching behavior also changes. ModelInput / 1MOutput / 1MBest for Sol$5$30Long-horizon coding, security, agents Terra$2.50$15High-volume production work Luna$1$6Fast, routine, low-cost tasks Sol’s $5/$30 matches GPT-5.5’s pricing. Terra is about 2x cheaper than GPT-5.5. Prompt caching now supports explicit cache breakpoints and a 30-minute minimum cache life. Cache writes cost 1.25x the uncached input rate. Cache reads keep the 90% discount. OpenAI also plans to run Sol on Cerebras hardware. It targets up to 750 tokens per second in July. Use Cases With Examples Long-horizon coding agents: Sol’s Terminal-Bench gains suit multi-step CLI automation. Example: an agent that plans, edits files, runs tests, then iterates. High-volume production: Terra fits chat features and document processing at scale. Example: summarizing thousands of support tickets each day at lower cost. Latency-sensitive apps: Luna suits autocomplete, routing, and simple extraction. Example: classifying inbound emails before a heavier model handles edge cases. Defensive security work: Sol targets vulnerability research and patching. Example: reviewing a codebase to find and fix a memory bug. Strengths and Open Questions Strengths Clear tiering across cost, speed, and intelligence New ultra subagent mode for complex, parallel work Reported state-of-the-art on Terminal-Bench 2.1 Token-efficiency gains on biology and cyber benchmarks A documented, layered safety stack Open questions Access is limited to about 20 partners at preview Public benchmark detail is partial until general availability Safeguards may block some legitimate dual-use security work Pricing sits above some open-weight competitors like GLM-5.2 Real-world latency for max and ultra is not yet public