GPT 5.6 Sol on VM0. OpenAI's newest flagship reasoning model
The flagship tier of OpenAI's GPT-5.6 preview generation. The strongest pick for agentic coding, deep reasoning and long computer-use loops on the OpenAI side.
400K tokens · Text / Vision / Code · Prompt cache
GPT 5.6 Sol is the top tier of OpenAI's GPT-5.6 preview generation and the model you reach for when the work needs both deep reasoning and reliable tool use: orchestrating multi-step agent loops, code edits that have to land first try, and computer-use workflows that span many GUI actions.
Vendor list price is $5 / $30 per 1M tokens with cached input at $0.50 / 1M. It sits at the ×2 credit tier on VM0 — the frontier-flagship price band — so the cost-effective pattern is to keep GPT 5.6 Terra or Claude Sonnet 4.6 as the everywhere-default and route only the hardest steps to Sol. GPT-5.6 is currently a preview family on VM0; GPT-5.5 remains the default OpenAI model until Sol graduates.
What is GPT 5.6 Sol?
July 2026 preview (top tier of the GPT-5.6 family) · Flagship tier of the GPT-5.6 family, above GPT 5.6 Terra and GPT 5.6 Luna. OpenAI's newest frontier model for agentic coding and reasoning.
GPT-5.6 is OpenAI's newest GPT generation, introduced in July 2026 as a three-tier family — Sol, Terra and Luna — rather than a single model. Sol is the flagship: the tier OpenAI positions for the hardest agentic-coding, reasoning and computer-use work. The 400K-token context window and the reasoning_effort parameter carry over from GPT-5.5 unchanged, so existing Codex agents drop in without rewrites.
Against GPT-5.5 (the previous flagship), Sol spends more compute per token on reasoning and, in OpenAI's preview materials, lands stronger first-attempt patches on multi-file refactors, fewer mis-routed tool calls on long loops, and gains on graduate-science reasoning and competition math. The trade-off is that it shares GPT-5.5's list price ($5 / $30 per 1M tokens) and the ×2 credit multiplier on VM0, which is why OpenAI frames Sol as the planner or escalation tier rather than the everywhere-default.
Because GPT-5.6 is a preview family, published numbers are early and will move. OpenAI has flagged training-data contamination on SWE-bench Verified across frontier models, so treat any absolute score as directional. The durable signal is the structured behaviour: tool-call accuracy, computer-use reliability and first-attempt patch quality, all of which the preview data shows improving over GPT-5.5.
What's notable about GPT 5.6 Sol
Headline architecture and capability features.
GPT 5.6 Sol keeps the 400K-token context window from the GPT-5 line, billed at standard input pricing across the entire window. It supports the reasoning_effort parameter, prompt caching where cached input bills at one-tenth the input rate ($0.50 / 1M) plus a cache-write charge ($6.25 / 1M), and the Responses API surface that the codex CLI uses by default. Tool-use, structured outputs and computer-use carry over from GPT-5.5. Inputs are multimodal across text, vision and code; there is no native image generation (use the Images API).
Specs at a glance
GPT 5.6 Sol benchmarks
Preview figures from OpenAI's GPT-5.6 Sol materials, shown against the public GPT-5.5 numbers. Treat all percentages as directional: the family is in preview and OpenAI has flagged training-data contamination on SWE-bench Verified across frontier models.
GPT 5.6 Sol pricing
Provider list price, per 1M tokens.
How GPT 5.6 Sol behaves in practice
Observed behaviour from production agent runs.
Tool routing
Lowest rate of mis-routed tool calls in the GPT-5.6 family. The gap versus GPT-5.5 shows up on hard edge cases: conditional tool selection, deeply nested arguments, and tool calls dispatched after long stretches of reasoning.
First-attempt code edits
Strongest patch quality in the family. The right pick when an agent has to modify code that must keep compiling and passing tests, especially when the patch spans multiple files. Preview SWE-bench Verified reflects this directly.
Computer use
More reliable than GPT-5.5 on multi-step GUI sequences, which is what the OSWorld delta captures. Reach for it when the agent drives a browser or desktop app over dozens of steps and a mid-run derailment is costly.
Speed
Slower than Terra and much slower than Luna. Around 70 tokens/sec at medium effort in early testing. Reserve it for the steps that genuinely need the extra reasoning depth and run lighter tiers in parallel.
Hallucination behaviour
Sol carries OpenAI's stricter calibration from the GPT-5 line and tends to admit uncertainty rather than confabulate, which is why teams keep paying the premium for high-stakes reasoning even as cheaper models catch up on benchmarks.
Best agent tasks for GPT 5.6 Sol
The orchestrator running a multi-tool plan
Use Sol as the planner that breaks a request into ten steps, dispatches each to a Terra- or Luna-tier sub-agent, and stitches the results back together. Running Sol only at the planner layer costs a fraction of running it end-to-end, with most of the quality preserved.
The first-try code edits that don't waste a CI run
Ask Sol to migrate a 50-file codebase from one ORM to another, refactor a tangled module, or apply a security fix across the repo. The patch applies cleanly on the first attempt more often than any other tier in the family — which is exactly what your CI bill reflects.
The computer-use agent that has to finish the workflow
When the agent drives a browser through a multi-step booking flow, a desktop app, or a legacy admin UI, Sol's stronger OSWorld score means fewer mid-run derailments and fewer human takeovers.
The hard-math or hard-science research step
Drop a competition-grade math set or a graduate physics derivation in and Sol works through it without the off-by-one slips you see on lighter tiers. AIME 2025 and GPQA Diamond pick up exactly this behaviour.
When to skip GPT 5.6 Sol
Skip Sol on high-volume routine work where GPT 5.6 Terra hits the same quality bar at half the credit cost, on latency-sensitive chat replies where GPT 5.6 Luna is much faster, and on bulk classification or extraction jobs where Luna is the cheaper supported option.
GPT 5.6 Sol vs other models
GPT 5.6 Sol vs GPT 5.6 Terra
Terra is the balanced workhorse of the GPT-5.6 family and the right pick for most agents. Promote to Sol only when Terra visibly fails on hard reasoning, long agentic loops or first-attempt code edits — usually as the orchestrator that delegates downward to Terra and Luna.
GPT 5.6 Sol vs GPT-5.5
Same role, newer generation. Sol is the preview successor to GPT-5.5 at the same list price and credit tier, with stronger preview scores on agentic coding and computer use. Until GPT-5.6 graduates from preview, GPT-5.5 stays the default OpenAI model on VM0; reach for Sol when you want the newest reasoning depth.
GPT 5.6 Sol vs Claude Opus 4.8
Peer flagships in different families: the high-stakes orchestrator you escalate to when the cheaper tier fails. Opus 4.8 has the 1M-token context window and Anthropic's safety profile; Sol has the Codex ecosystem and OpenAI's computer-use scores. Pick by which framework your existing agents target.
Bottom line: should you use GPT 5.6 Sol?
GPT 5.6 Sol is the newest escalation tier on the OpenAI side. Default to Terra; promote to Sol only on the specific steps where Terra visibly falls short, and expect preview scores to keep moving.
Frequently asked questions
What is GPT 5.6 Sol's context window?
400,000 tokens, with up to 128K tokens of output per response. The full window bills at standard rates.
Is GPT-5.6 generally available on VM0?
It is in preview. All three tiers — Sol, Terra and Luna — are selectable, but GPT-5.5 remains the default OpenAI model until the GPT-5.6 family graduates. Preview benchmark numbers are early and will move.
When should I pick Sol over GPT 5.6 Terra?
When (a) the agent is the planner and decisions cascade, (b) the run is long enough that Terra starts mis-routing tool calls, or (c) the output must apply cleanly on the first attempt — code edits, structured payloads, computer-use workflows.
Does GPT 5.6 Sol support prompt caching?
Yes. Cached input bills at $0.50 per 1M tokens — a 10× discount on the cached portion — with a cache-write charge of $6.25 per 1M. Worth using whenever your system prompt or tool schema is stable across calls.
What framework does GPT 5.6 Sol use on VM0?
Codex. VM0 routes GPT-5.6 through the Codex framework's Responses API surface, which is what the codex CLI uses by default. Claude Code-framework agents are not compatible with GPT-5 models on VM0.
Alternatives
Using GPT 5.6 Sol on VM0
Two ways to access GPT 5.6 Sol on VM0
VM0 supports GPT 5.6 Sol as a Built-in model billed in VM0 credits, and through bring-your-own with a OpenAI API key. The Built-in path uses VM0 Managed routing and the credit multiplier explained below; the bring-your-own path bills you directly with the upstream vendor and skips the VM0 credit conversion entirely.
VM0's recommendation
VM0 positions GPT 5.6 Sol as a core agent model, recommended alongside Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 for the steps that drive the actual outcome of an agent run. These are the models we'd pick for the orchestrator role, for code-touching agents, and for any step where a wrong answer is expensive.
Credits and the ×2 multiplier
Every Built-in model on VM0 is priced as a multiple of Claude Sonnet 4.6, which sits at the ×1 credit baseline. GPT 5.6 Sol bills at ×2 credits. The multiplier is what shows up on your VM0 invoice; the vendor list price in the pricing table above is what the upstream provider charges before VM0 converts it into credits.
GPT 5.6 Sol bills at ×2, which means a step here costs 2× the credits of an equivalent step on Sonnet 4.6 (the ×1 baseline). It's a premium tier on VM0, so the cost-effective pattern is to default to a cheaper model and route only the steps that genuinely need the extra reasoning depth to GPT 5.6 Sol.
Available on VM0 since July 2026 (preview).