GPT-5.6

On June 26, 2026, OpenAI began a limited preview of the GPT-5.6 series—Sol (flagship), Terra (balanced), and Luna (fast and affordable)—for a small group of trusted partners at the U.S. administration's request, ahead of broader availability in the coming weeks. The post describes GPT-5.6 Sol launching with OpenAI's mo It reports that Terra is competitive with GPT-5.5 while being 2x cheaper, and that Luna brings strong capability at OpenAI's lowest cost tier. OpenAI frames the government pre-coordination as a short-term step tied to the cyber Executive Order framework and signals an expanded suite of evaluation results to be shared w

OpenAI published a builder's guide to GPT-5.6 on August 13, 2026, featuring technical lessons from startups using the model family in production. The guide emphasizes price-performance improvements, showing GPT-5.6 Sol, Terra, and Luna on the Artificial Analysis Coding Index alongside GPT-5.5, Claude Fable 5, Claude Op Concrete developer guidance includes reasoning-effort tuning: GPT-5.6 Sol at "low" reasoning outperformed GPT-5.5 at "high" reasoning on Agents' Last Exam with a constant harness. Model selection advice notes that Luna and Terra can match GPT-5.4 and 5.5 performance with more test-time compute at significantly lower co

OpenAI's August 2026 GPT-5.6 system card describes a rollout that replaces GPT-5.5 Instant for Free and Go users, while Plus and Pro get an updated GPT-5.6 Sol with an effort slider; the August versions of Sol and Luna are distinguished from the July versions still served in Codex and ChatGPT Work. Under the Preparedne The report adds new evaluation surfaces not previously standard, including dedicated Under-18 (U18) safety benchmarks, dynamic mental-health benchmarks with adversarial user simulations, and expanded jailbreak, prompt-injection, and vision robustness testing. Safety evaluations are run at the lowest reasoning setting u

The Agent Report published a detailed third-party analysis on July 13, 2026, breaking down the GPT-5.6 family launch. The piece confirms the three-tier architecture (Sol at $5/$30, Terra at $2.50/$15, Luna at $1/$6 per million tokens) and reports that Sol leads on agentic benchmarks including Terminal-Bench 2.1 (88.8%) The analysis frames GPT-5.6 as a structural shift: OpenAI replaced the single-model launch with three durable capability tiers evolving on independent cadences, plus two new levers (max reasoning effort and ultra mode for multi-agent parallelism). Explicit model IDs are gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna, wit

CodeRabbit published a July 9, 2026 engineering review of GPT-5.6 Sol and Terra focused on coding agents and code review. The main takeaway is that Sol is stronger on long-horizon coding work: it follows through across messy repo tasks, takes long checklists seriously, and finds more review issues in CodeRabbit's harne The review notes that Sol does not displace every frontier model: Fable 5 still feels stronger on architectural judgment and planning taste, and Sonnet 5 has a cleaner comment-quality story in some review workflows. The key release changes for engineering teams are: Sol for long-horizon coding, Terra for cheaper scoped

OpenAI announced general availability of the GPT-5.6 family on July 9, 2026, following a limited preview that began June 26. The family includes three tiers: Sol as the new flagship, Terra as a balanced everyday-work model, and Luna as the most cost-efficient option. The post emphasizes that GPT-5.6 Sol is trained to d Key technical details include a new "ultra" capability setting that coordinates multiple agents across parallel workstreams for demanding work, and improved computer-use and design judgment. On Agents' Last Exam across 55 fields, GPT-5.6 Sol reaches 53.6, beating Claude Fable 5 (adaptive reasoning) by 13.1 points, and

Artificial Analysis published independent benchmark results on July 9, 2026, noting it supported OpenAI with pre-release evaluation of GPT-5.6 Sol, Terra, and Luna. GPT-5.6 Sol (max) scores 59 on the Artificial Analysis Intelligence Index—1 point below Claude Fable 5 (max)—at approximately one third of the cost ($1.04 The analysis reports that across reasoning efforts, each GPT-5.6 model pushes past GPT-5.5 on the Pareto frontier (excluding non-reasoning), with Luna and Sol consistently on the frontier ahead of Terra—meaning a Luna or Sol effort level is more intelligent at no extra cost, or equally intelligent at lower cost, than T

BenchLM.ai's profile for GPT-5.6 Sol lists it as a proprietary reasoning model released July 9, 2026, with a 1.05M-token context window and aggregated API pricing of $5 per million input tokens, $30 per million output tokens, and $0.50 per million cached input tokens (a $17.50 blended rate). It reports a capability sco The same profile notes a measured throughput of roughly 64 tokens/second with a first-token latency around 137.75s against a field median of 89 tok/s, and calls out that only 25 source-displayable benchmark rows are published, leaving several tracked slots empty. These figures give developers a concrete, third-party co