GPT-6.1 Sol is OpenAI's upgrade to GPT-6 Sol, designed for real codebase modifications, complex document analysis, and multi-step business workflows. In official release evaluations, it approaches GPT-6 Astra on multiple programming and professional tasks, while improving the previous generation Sol's factual accuracy and task completion capabilities. It is suited for development and knowledge work that requires strong reasoning while controlling the cost of the overall task.
Point the Base URL to api.acedata.cloud, configure your platform API key and the model ID below, and use your compatible SDK or client.
api.acedata.cloudgpt-6.1-solimport os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["ACEDATACLOUD_API_KEY"],
base_url="https://api.acedata.cloud/v1",
)
response = client.responses.create(
model="gpt-6.1-sol",
input="Hello!",
)
print(response.output_text)Choose an available protocol for this model. OpenAI SDK uses a Base URL ending in /v1; Anthropic SDK uses the root URL. See each guide for protocol-specific parameters, tools and response formats.
Clarify capacity, inputs and outputs, and invocation methods before choosing a model.
Capacity and native capabilities are based on publicly available model information; actual request formats, available tools, and limitations are subject to the API documentation for this platform. Context budget and maximum output should be planned together.
Learn what gpt-6.1-sol can bring to your work.
The official DeepSWE 1.1 evaluation focuses on long-horizon software engineering tasks in real codebases, where GPT-6.1 Sol achieves performance close to Astra. For developers, it is better suited to providing the problem description, relevant modules, and constraints together, allowing the model to trace call relationships, identify defects, and develop cross-file modification plans.
The official GDP.pdf evaluation covers tables, charts, diagrams, and small text in professional PDFs, demonstrating its document-understanding capabilities. It can be used to compare specifications, analyze reports, or extract business fields; applications should provide text or image materials as required by the API and retain sources that can be checked against the original text.
It is suited to breaking business goals into connected steps rather than merely producing a piece of advice. Responses can be used to organize inputs and results for multi-step tool tasks; applications need to define tools, manage permissions, and retain execution records. After the model proposes actions, actual tool returns should be used to determine whether queries, writes, and verifications have been completed.
Start with specific tasks to find where the model can be effective.
Provide a specific repository version, reproduction information, and relevant modules, then have the model analyze the scope of changes and dependencies before organizing a patch or refactoring plan. Compared with asking about a single function, these tasks make better use of its ability to understand engineering context; final results should still be validated through the project's actual execution and review.
For specifications, contract drafts, or research reports, ask the model to answer specific questions and list supporting evidence, or organize table contents by field. Distinguishing facts, inferences, and missing information in the materials helps turn long-document analysis into business results that are easier to review.
After the application provides controlled function tools, it can organize continuous workflows such as querying, checking, filling out forms, and updating records. First define the conditions the final record must meet, then feed tool results back to the model; do not mistake a “model-planned update” for a completed system write.
Choose based on task complexity, input materials, and expected results.
The main upgrades in 6.1 Sol are code, professional documents, and complex workflows; the official report also notes improved factual accuracy and lower cache read costs. When migrating existing Sol applications, in addition to comparing output quality, check reasoning levels and tool protocols; do not assume requests are fully compatible simply by replacing the model string.
For the most difficult end-to-end research and demanding tasks, you can continue comparing Astra; for clearly bounded classification, extraction, and high-frequency subtasks, you can compare Luna. During evaluation, observe completion quality, output usage, retries, and human revision costs as well; model benchmark scores cannot be directly equated with your production results.
Based on the characteristics of gpt-6.1-sol, first validate a small task whose results can be checked.
You can ask directly: Compare these three modules, the failure logs, and the interface contract, and identify the root cause of the behavior change after the upgrade. List the evidence first, then provide the minimal modification plan and regression checks, without expanding the feature scope.
Provide the repository version, call chain, and actual failure logs; validate interface compatibility and test results separately.
Use the full model ID gpt-6.1-sol, first confirm the public request format and available parameters on the API page, then connect your application. Retain result parsing, exception handling, and relevant evidence, and evaluate whether it is suitable for continued use with the same set of real samples.
Before formal use, understand output quality and capability scope.
Answers to common questions about using gpt-6.1-sol.
Key official release highlights include improvements in real codebase tasks, complex PDF analysis, and multi-step business workflows, as well as better factual accuracy and lower official cached input pricing. Similar capacity does not mean calls are fully compatible; tool requests and reasoning tiers need to be verified for the new model.
No. The official model documentation separately lists a maximum input of 922,000 tokens and a maximum output of 128,000 tokens; the 1,050,000-token context must accommodate both relevant input and output. Actual requests must also meet this platform's API limits, and images, tool content, and history must be included in the budget.
Tool workflows use Responses, with function definitions, call results, and execution information configured and returned according to the platform documentation. The latest official model documentation explicitly states that Chat Completions for gpt-6.1-sol does not support tool calling; support cannot be inferred merely because shared requests include a tools field. Standard text-and-image Q&A can use Chat Completions.
The platform guide recommends starting with low, comparing medium, high, or xhigh using the same set of tasks, then evaluating max for complex tasks that genuinely require it. Higher tiers should be determined by completion quality, while also considering reasoning usage and latency. Do not continue using none or minimal, and set reasoning_effort or reasoning according to the selected API.
This platform's billing rules distinguish among standard input, cached input reads, cached input writes, and output; when prompt_tokens exceeds 272,000, the long-context tier applies, with input and cache-related rates at 2 times the corresponding base tier and output at 1.5 times. Amounts are subject to the Pricing page and selected plan; repeated text does not necessarily mean the cache has been hit.
Choose a development approach or application suited to this model and follow its setup guide.
↗OpenAI Chat Completions API
SDK installation, authentication and examples↗OpenAI Responses API
SDK installation, authentication and examples
↗Using Ace Data Cloud with Cherry Studio
Chat client setup · Base URL, API key and model
↗Using Ace Data Cloud with ChatBox
Chat client setup · Base URL, API key and modelThis is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs