GPT-5.6 Sol

GPT-5.6 Sol

Quick answer: GPT-5.6 Sol is OpenAI's highest-capability tier in the GPT-5.6 family, released July 9, 2026. It scores 94.6% on GPQA Diamond, 88.8% on TerminalBench 2.1, 97.5% on ARC-AGI, and 52.7% on Agent's Last Exam — the top result in the GPT-5.6 lineup. Priced at $5 input / $30 output per 1M tokens with a 1.1M-token context window.

At a Glance

Where GPT-5.6 Sol leads

• Highest intelligence tier of the GPT-5.6 family: AA Intelligence Index 58.9, Coding Agent Index 80
• 94.6% GPQA Diamond — among the strongest scientific reasoning scores reported
• 88.8% TerminalBench 2.1 — state-of-the-art for command-line agentic workflows
• 97.5% ARC-AGI — near-perfect abstract reasoning
• 90.84% BrowseComp — top-tier web research and retrieval
• max and ultra reasoning effort modes for demanding multi-step tasks
• 1.1M-token context window — supports whole-codebase and long-document analysis

Where it lags

• Most expensive variant in the GPT-5.6 family at $5/$30 per 1M tokens
• Closed API model; no open weights available
• ultra multi-agent mode requires Pro or Enterprise plan access

Best for: Frontier-difficulty agentic coding, cybersecurity analysis, scientific reasoning, and long-horizon professional tasks where maximum intelligence is required.

What GPT-5.6 Sol Is

GPT-5.6 Sol is the flagship tier of OpenAI's GPT-5.6 model family — a generational step following GPT-5.5, built around delivering more useful work per token. The GPT-5.6 family introduced a durable three-tier naming scheme: Sol (highest capability), Terra (balanced), and Luna (most efficient), with each tier designed to advance independently as OpenAI updates the family.

Sol is distinguished from its siblings primarily by its reasoning effort ceiling. It supports max mode — where the model reasons longer, explores alternatives, and revises before responding — and ultra mode, which orchestrates four parallel subagents by default. ultra makes GPT-5.6 Sol the first OpenAI model to natively expose a multi-agent mode at the API level, delivering faster completion on demanding multi-step tasks at the cost of higher token usage.

For agent builders, Sol's benchmark profile translates directly to task performance: the combination of 88.8% TerminalBench and 97.5% ARC-AGI indicates strong generalisation across structured and novel task types, while the 64.6% SWE-Bench Pro score positions it as a competitive choice for long-horizon software engineering workflows.

Specifications

Pricing

Cache writes are billed at 1.25× the uncached input rate; cache reads receive a 90% discount. Pricing per OpenAI pricing page as of July 2026.

Context Window

GPT-5.6 Sol has a ~1.1M-token context window — roughly 800–900 pages of text in a single request. This is sufficient to load entire mid-size codebases, lengthy legal documents, or multi-session conversation histories without chunking. Long-context retrieval is confirmed at 91.5% on MRCR v2 (8-needle, 256K–512K range).

Public Benchmark Scores

Scores are reported by OpenAI or measured by Benchgen and shown for context. Figures depend on harness, tools, and effort settings — see the source for methodology.

GPT-5.6 Sol vs Alternatives

Sol sits at the top of the GPT-5.6 price-performance curve: it costs 2× Terra but delivers meaningfully stronger results on agentic and reasoning tasks where intelligence ceiling matters more than throughput.

How GPT-5.6 Sol Performs on Real Agent Tasks

Public benchmarks establish Sol's ceiling: a 97.5% ARC-AGI score and 88.8% TerminalBench place it at the frontier for abstract reasoning and terminal-based coding tasks respectively. The Agent's Last Exam score of 52.7% — the highest in its family — reflects Sol's advantage on tasks requiring sustained multi-step planning and tool use.

For agent builders evaluating Sol against Terra or Luna, the practical question is not raw capability but task difficulty. On straightforward automation and content tasks, Terra and Luna close the gap significantly at 40–80% lower cost. Sol's advantage compounds on tasks that push near the model's limit: novel debugging scenarios, security analysis, and multi-tool orchestration where the max and ultra reasoning modes add concrete value.

Use GPT-5.6 Sol via API

from openai import OpenAI

client = OpenAI(api_key="YOUR_API_KEY")

response = client.chat.completions.create(
    model="gpt-5.6-sol",
    messages=[{"role": "user", "content": "Analyse this codebase for security vulnerabilities..."}],
)
print(response.choices[0].message.content)

Related Models & Benchmarks

• Compare: GPT-5.6 Terra · GPT-5.6 Luna · GPT-5.5 · Grok 4
• Benchmarks: GPQA Diamond · ARC-AGI · TerminalBench · SWE-Bench Pro · Agent's Last Exam

Frequently Asked Questions

Specs and scores sourced from OpenAI's official GPT-5.6 announcement (July 9, 2026) and Benchgen evaluations; third-party benchmark scores attributed inline. Pricing cited to the OpenAI pricing page. Last updated 2026-07-21.