GPT-5.6 Sol
Complex professional work, coding, research, computer use and demanding agentic workflows.
gpt-5.6-sol1,050,000 context · 128,000 max output
$4.00input$20.00output / 1M
Official model page ↗
MODEL INTELLIGENCE
GPT-5.6
model family reference.
GPT-5.6 model family reference: Sol, Terra and Luna pricing, 1.05M context window, API aliases, reasoning modes, tools, availability and current primary-source updates.
3tracked signals
1primary sources
September 21, 2026latest tracked
MODEL REFERENCE
GPT-5.6 at a glance.
GPT-5.6 is OpenAI’s 2026 model family for production reasoning, coding, knowledge work and agentic workflows. The family has three durable tiers: Sol for flagship capability, Terra for a balance of intelligence and cost, and Luna for efficient high-volume work.
The unsuffixed API alias gpt-5.6 routes to GPT-5.6 Sol. That distinction matters: “GPT-5.6” can refer to the family in product discussions, while the API alias resolves to the Sol tier.
Introduced July 9, 2026 · Text and image input · text output · Last verified 2026-11-10
FAMILYSol · Terra · Luna
API ALIASgpt-5.6 → Sol
CONTEXT WINDOW1,050,000tokens across family
MAX OUTPUT128,000tokens across family
KNOWLEDGE CUTOFFFeb 16, 2026
DEFAULT REASONINGMediumnone → max supported
FAMILY MAP
Three GPT-5.6 tiers, three cost profiles.
The models share the same large context and output limits, but target different quality, latency and cost envelopes.
GPT-5.6 Terra
Everyday production workloads that need strong reasoning at a lower unit cost than Sol.
gpt-5.6-terra1,050,000 context · 128,000 max output
$2.00input$12.00output / 1M
Official model page ↗
GPT-5.6 Luna
Cost-sensitive, high-volume and latency-conscious workloads.
gpt-5.6-luna1,050,000 context · 128,000 max output
$0.20input$1.20output / 1M
Official model page ↗
Current Standard API pricing per 1M text tokens. Prompts above 272K input tokens use higher long-context rates. GPT-5.6 Sol is currently on promotional pricing listed by OpenAI through at least November 21, 2026.
WHAT GPT-5.6 IS
A production reasoning family, not one single model.
OpenAI introduced GPT-5.6 as a family designed to improve performance per dollar across complex work. Sol is the flagship tier and roughly takes the place of the unsuffixed flagship position from earlier GPT-5 families. Terra is the balanced tier, while Luna is optimized for cost-sensitive throughput.
For developers, that means model selection can happen inside one generation without changing the overall API architecture. You can reserve Sol for the tasks where additional reasoning materially changes the result, move repeatable production work to Terra, and use Luna where volume and unit economics dominate.
GPT-5.6 is no longer OpenAI’s newest generation—GPT-6 now occupies that position—but GPT-5.6 remains a current production family with documented API support and distinct lower-cost tiers.
CAPABILITIES
What GPT-5.6 is designed to do well.
Adjustable reasoning effort
GPT-5.6 supports none, low, medium, high, xhigh and max reasoning effort. OpenAI recommends medium as a balanced starting point, then measuring whether higher effort creates a real quality gain.
Complex software work
OpenAI positions GPT-5.6 for coding and production workflows, with stronger token efficiency and improved frontend design judgment compared with earlier models.
Tool-oriented workflows
The Responses API supports programmatic tool calling, persisted reasoning and agent workflows. Multi-agent orchestration is available in beta across GPT-5.6 models.
Text + image input
The family accepts image input and text input, with text output. Current model pages do not list native audio or video support for these models.
1.05M-token context
All three variants list a 1,050,000-token context window and up to 128,000 output tokens, enabling large repositories, document sets and long-running workflows.
Tiered price-performance
Sol, Terra and Luna let the same application route different task classes to different cost envelopes instead of treating every request as a flagship-model request.
API & TOOLS
GPT-5.6 in the OpenAI API.
For reasoning and tool-calling workflows, OpenAI recommends the Responses API rather than treating the model as a simple text-completion endpoint.
RESPONSES APIRecommended for reasoning + tools
Use the Responses API for multi-turn reasoning, function calling and agent workflows.
PROGRAMMATIC TOOL CALLINGModel-written coordination logic
GPT-5.6 can coordinate eligible tools and intermediate results in a hosted runtime.
MULTI-AGENTBeta across GPT-5.6
A root model can delegate parallel work to subagents and synthesize the outputs in one workflow.
TOOLSWeb, files, code, shell, computer use, MCP
OpenAI’s current Sol model page lists web search, file search, code interpreter, hosted shell, computer use, MCP, tool search and more.
STRUCTURED OUTPUTSSupported
Useful for production applications that need machine-readable model results rather than free-form prose.
FINE-TUNINGNot currently supported
OpenAI’s current GPT-5.6 Sol model page does not list fine-tuning support.
REASONING MODES
Standard effort, max effort and Pro mode.
Reasoning effort controls how much reasoning the model applies within a request. GPT-5.6 supports six levels from none through max, and defaults to medium. Lower settings generally favor latency and cost; higher settings are appropriate only when the workload shows a measurable quality benefit.
Pro mode is separate from reasoning effort. Instead of switching to a separate GPT-5.6 Pro model slug, OpenAI documents reasoning.mode: "pro" on the selected GPT-5.6 model. Pro mode performs more model work and is intended for difficult, quality-first tasks that can tolerate higher latency and token usage.
GPT-5.6 also changed reasoning persistence: the family can reuse compatible reasoning across turns within the same GPT-5.6 family. That can matter in long, tool-heavy workflows where goals and assumptions remain stable across multiple calls.
PRICING
GPT-5.6 API pricing.
Current Standard text-token rates per 1M tokens. Cached input is substantially cheaper, while very long prompts use higher rates.
Long-context pricing
OpenAI’s current model pages state that prompts above 272K input tokens are priced at 2× the input rate and 1.5× the output rate for the full request. Cache writes are billed above ordinary uncached input, while cache reads receive a large discount.
Sol promotional pricing
OpenAI currently labels GPT-5.6 Sol’s $4 / $20 rates as promotional pricing available at least through November 21, 2026. Recheck the official model page before building a long-term cost model.
AVAILABILITY
Where GPT-5.6 is available.
Sol, Terra and Luna
All three GPT-5.6 tiers are available through the API, with model-specific rate limits and tool support.
GPT-5.6 across coding workflows
OpenAI’s current help documentation lists Terra for Free and Go, with Sol, Terra and Luna available to higher plans in Codex.
Tier choice on paid plans
Current help documentation lists Sol, Terra and Luna in Work for Plus, Pro, Business and Enterprise users.
Sol and Luna remain product models
OpenAI has continued to update Sol for paid ChatGPT use and Luna for the default Free/Go experience, while availability varies by plan and surface.
WHICH MODEL?
Choose GPT-5.6 by workload, not by name.
Quality-first complex work
Use Sol for difficult coding, professional analysis, computer use, research and agentic tasks where additional reasoning changes the outcome enough to justify the higher unit cost.
Balanced production workloads
Use Terra when the application needs strong reasoning but does not consistently benefit from Sol’s extra capability. It is the natural tier to evaluate for everyday production routing.
High-volume efficiency
Use Luna for classification, extraction, focused transformations and other repeatable workloads where throughput and price matter more than maximum reasoning depth.
When you need the newer frontier
GPT-6 is OpenAI’s newer generation. For new high-capability systems, compare GPT-5.6 against GPT-6 rather than assuming the older family remains the best quality choice.
Open GPT-6 reference ↗MIGRATION
Moving from GPT-5.5 or GPT-5.4.
OpenAI recommends beginning with your current reasoning setting rather than automatically raising effort on GPT-5.6. Test the same setting and one level lower on representative workloads, then compare quality, latency and token use.
For applications that use tools or multi-step reasoning, migration is also a chance to move toward the Responses API and evaluate persisted reasoning, Programmatic Tool Calling and beta multi-agent orchestration. Do not change model, API architecture and agent permissions all at once unless you can isolate the effect of each change in evaluation.
Older GPT-5 and o3 snapshots are also on a published deprecation path, with OpenAI listing GPT-5.6 Sol, Terra and Luna as replacement directions for several older GPT-5 tiers.
LIMITS & CAVEATS
What to keep in mind.
01GPT-6 is newer.
GPT-5.6 remains useful, but “latest” and “highest capability” now refer to the GPT-6 generation for many workloads.
02Long context is not free context.
A 1.05M-token window is a capacity limit, not a reason to send unnecessary data. Very long prompts cost more and can increase latency.
03Benchmarks are workload-specific.
OpenAI’s launch evaluations are useful evidence, but production routing should still be based on your own task distribution and acceptance criteria.
04Tool access changes the risk model.
Agents that can browse, execute code, use MCP or call business systems need permission, sandboxing and approval controls around the model.
05Pricing can change.
Sol is explicitly on promotional pricing at the time of verification, so production cost models should reference the current official model page.
06Fine-tuning is not listed.
The current Sol model page lists fine-tuning as unsupported; use prompting, tools, retrieval and model routing where appropriate instead of assuming a fine-tuned GPT-5.6 endpoint exists.
OFFICIAL SOURCES
OpenAI · GPT-5.6 launch↗OpenAI API · GPT-5.6 model guidance↗OpenAI API · GPT-5.6 Sol↗OpenAI API · GPT-5.6 Terra↗OpenAI API · GPT-5.6 Luna↗OpenAI API · Multi-agent↗OpenAI Help · GPT-5.6 availability↗QUICK ANSWERS
What is GPT-5.6?
GPT-5.6 is an OpenAI model family introduced in 2026. It includes Sol for flagship capability, Terra for a balance of intelligence and cost, and Luna for efficient high-volume workloads.
What does the gpt-5.6 API alias use?
OpenAI states that the gpt-5.6 alias routes requests to GPT-5.6 Sol.
What is the GPT-5.6 context window?
GPT-5.6 Sol, Terra and Luna each list a 1,050,000-token context window and up to 128,000 output tokens.
How much does GPT-5.6 cost?
Current Standard API pricing is $4 input / $20 output per 1M tokens for Sol, $2 / $12 for Terra, and $0.20 / $1.20 for Luna. Cached-input rates are lower, and long prompts above 272K input tokens use higher rates.
Which GPT-5.6 model should I use?
Use Sol when quality and complex reasoning matter most, Terra when you want a stronger cost-performance balance, and Luna when throughput and cost are primary constraints.
Does GPT-5.6 support reasoning effort?
Yes. OpenAI documents none, low, medium, high, xhigh and max reasoning effort for the GPT-5.6 family, with medium as the default.
Does GPT-5.6 support tools and agents?
Yes. OpenAI documents tool use through the Responses API, including web search, file search, code interpreter, hosted shell, computer use, MCP and beta multi-agent orchestration.
Is GPT-5.6 still relevant after GPT-6?
Yes. GPT-6 is the newer generation, but GPT-5.6 remains an available production family across OpenAI products and the API, with lower-cost Terra and Luna tiers that can be useful for cost-sensitive workloads.
LATEST DEVELOPMENT
V7 cuts costs 78% while boosting accuracy with GPT-5.6 Luna
Using GPT-5.6, V7 turns scattered company files into context agents can use to complete complex, source-linked work.
Open latest signal ↗WHAT CHANGED
GPT-5.6: what changed.
No post-baseline factual changes recorded for GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna since 2026-09-27.
VERIFIED BASELINE2026-09-27Current facts remain aligned with the SXF ledger.
VERIFIED CHANGE HISTORY
GPT-5.6 verified history.
GPT-5.6 Terra
Baseline verified: 1,050,000 context · 128,000 max output · $2.00 input / $0.20 cached / $12.00 output per 1 million tokens
Official evidence ↗GPT-5.6 Sol
Baseline verified: 1,050,000 context · 128,000 max output · $4.00 input / $0.40 cached / $20.00 output per 1 million tokens
Official evidence ↗GPT-5.6 Luna
Baseline verified: 1,050,000 context · 128,000 max output · $0.20 input / $0.02 cached / $1.20 output per 1 million tokens
Official evidence ↗Append-only SXF ledger · 3 verified events · chain head 5cb61a0bb057…
EVALUATION EVIDENCE
GPT-5.6 Terra benchmark observations.
independentArtificial Analysis
Artificial Analysis Intelligence Index 4.3.2
42 pointscomposite intelligence · reasoning max
Independent source ↗independentArtificial Analysis
AA-Briefcase 1.1
1,333 Eloagentic knowledge work · reasoning max
Independent source ↗independentArtificial Analysis
GDPval-AA 2.1
1,453 Eloprofessional knowledge work · reasoning max
Independent source ↗independentArtificial Analysis
AutomationBench-AA
60%business automation · reasoning max
Independent source ↗independentArtificial Analysis
Terminal-Bench 4.0
35%agentic coding · reasoning max
Independent source ↗independentArtificial Analysis · under review
SciCode
55%scientific coding · reasoning max
Evaluator labels this benchmark under review.
Independent source ↗independentArtificial Analysis
Humanity's Last Exam
43%multidisciplinary reasoning · reasoning max
Independent source ↗independentArtificial Analysis · under review
CritPt
30%frontier science reasoning · reasoning max
Evaluator labels this benchmark under review.
Independent source ↗independentArtificial Analysis
AA-LCR 1.1
83%long context reasoning · reasoning max
Independent source ↗Source-verified observations · scores depend on evaluation settings. Not a universal model ranking.