According to ChainCatcher, on June 28, OpenAI released a preview of its GPT-5.6 series, comprising three models: Sol (flagship), Terra (balanced), and Luna (fast, low-cost). Sol achieved 88.8% on Terminal-Bench 2.1 command-line workflow tests, reaching 91.9% in Ultra mode with sub-agent collaboration—surpassing GPT-5.5's 83.4% and Claude Fable 5's 88.0%. Terra delivers near-GPT-5.5 performance at half the price.
Sol is priced at $5 per million input tokens and $30 per output token, with prompt caching support to reduce subsequent call costs. Currently available to Codex and trusted API partners only.