GPT-5.6 Terra

OpenAI's June 26, 2026 launch post announces a limited preview of the GPT-5.6 series: Sol as the flagship, Terra as a balanced model for everyday work, and Luna as the fast, affordable tier. The post explicitly states Terra has "competitive performance to GPT-5.5 while being 2x cheaper," making this the authoritative u The post frames the preview as a deliberate, coordinated rollout tied to ongoing engagement with the U.S. government on a cyber Executive Order framework, and says the company will continue testing with partners as it works toward broader availability. It positions Sol as the strongest model in the series and previews

A DataCamp benchmark overview of Z.ai's GLM-5.3-Flash cites GPT-5.6 Terra in its Terminal-Bench 2.1 comparison, giving Terra a score of 87.4 ahead of Claude Opus 4.8 at 85.0 and GLM-5.3-Flash at 84.3. That puts the model in the near-frontier band on coding and agentic tasks, providing an independent data point for deve The same article frames the broader budget-tier landscape by pairing GLM-5.3-Flash (a 320B-A18B MoE with a 1M-token context, multimodal input, and a blended price around $0.10 per 1M tokens) against Qwen3.8-Flash-Next, which launched the same day. For Venice AI customers evaluating GPT-5.6 Terra on cost-versus-capabili

LayerLens's GPT-5.6 benchmark review covers Sol, Terra, and Luna across 12 benchmarks and reports a roughly 5x price spread from $1 to $5 per million input tokens across the tiers. For Terra specifically, the review cites pricing of $2.50/$15 per 1M input/output tokens and reports Terra scoring within three points of S The review frames the family's shared architecture with three capability tiers on the same API surface, and notes that Sol adds an "ultra" multi-agent mode coordinating four parallel sub-agents that pushes BrowseComp to 92.2% at 4x token cost. It also reports Sol's 1.05M-token context window holding at 91.5% on MRCR 8-

On 26 June 2026, OpenAI introduced the GPT-5.6 generation as a restricted preview containing three models: Sol as flagship, Terra as the capability/cost balance, and Luna as the fastest, least expensive tier. The piece notes the API identifiers `gpt-5.6-sol`, `gpt-5.6-terra`, and `gpt-5.6-luna`, with the shorter `gpt-5 The article distinguishes the two announcement stages: the June preview that established the family's shape, pricing posture, and safety framing, and the 9 July GA that expanded access, clarified how each tier appears across OpenAI's products, and added broader evaluations. It warns readers not to confuse model tiers w

The LLM Stats scorecard page for GPT-5.6 Terra places the model at an overall composite rank of 16, with a blended price of $2.48 per million tokens. It shows category standings where Terra rates as Good in Coding (8 of 267), Search (8 of 90), and Reasoning (16 of 363), Average in Chat, 3D, Long Context, SVG, Games, To The page lists per-benchmark scores sourced from the model's scorecard, paper, or official blog posts, including Connectors at 1.00/1 (rank 2) and HealthBench Consensus at 0.95/1 (rank 3). It frames Terra's pricing positioning between GPT OSS 120B ($0.043) and GPT-6 Astra ($11.9), with scores compared on a log-scale co

CodeRabbit's engineering review evaluates GPT-5.6 Sol and Terra for coding agents and code-review workflows. It positions Terra as the cheaper lane to test for scoped work, emphasizing cost-per-solved-task alongside token price, while recommending Sol for long-horizon coding tasks that require persistence across files, On benchmark comparisons, the piece reports Sol finding more review issues than Terra but notes that filtering has to earn user trust on flagged findings. It benchmarks Sol and Terra against competing frontier models, finding Sol does not erase rivals but wins on practical follow-through, while Terra remains a lower-co

OpenAI has previewed its GPT-5.6 model family in three variants — Sol, Terra, and Luna — per VentureBeat's June 26, 2026 report. Sol is described as the most capable, targeting hard problems such as complex coding and security research; Terra is aimed at high-volume business workloads including customer support, intern The rollout is initially restricted to roughly 20 organizations under a U.S. government-coordinated preview, following a June 2, 2026 executive order directing federal agencies to benchmark and assess new frontier models ahead of wider release. OpenAI states it previewed its plans with the government and is honoring a

The explainx.ai explainer documents the June 26, 2026 OpenAI preview of the GPT-5.6 family—Sol (flagship), Terra (balanced daily work), and Luna (fast, affordable volume)—and the July 9, 2026 rollout across ChatGPT, Codex, and the API. Terra was positioned as a balanced tier at $2.50 input / $15 output per 1M tokens, w Terminal-Bench 2.1 scores reported in the explainer place Sol Ultra at 91.9%, Sol at 88.8%, Mythos 5 at 84.3%, and Fable 5 at 83.4%, with OpenAI claiming that no GPT-5.6 tier crosses the Cyber Critical threshold. The post also notes that U.S. authorities requested a vetted-partner start that OpenAI opposes as a long-te