JSON that looks cut off can mean a token limit, invalid syntax or a missing field. Check the full Chat Completions response before changing your prompt. DeepSeek and other compatible providers can differ; use their documented response fields.
The checker runs in your browser without an account or model call. It does not guess missing content or repair a response. Use a non-streaming body; if you have only the assistant text, select JSON-only mode.
Check the request before the output
If your client recorded a non-successful HTTP status or an error object, diagnose that failure first. An authentication error, exhausted allowance or server failure is not a completed assistant answer. The local checker accepts an optional HTTP status; it cannot recover a status from the text alone.
Read finish_reason, not just the last character
For the first answer, inspect choices[0].finish_reason. In OpenAI's Chat Completions reference, the following values have distinct meanings:
| Reported reason | Interpretation | Next step |
|---|---|---|
length | The requested output-token limit was reached. Even parseable JSON may omit part of the requested task. | Reduce the output or use a model and host with a suitable supported limit. Do not fabricate the missing section. |
stop | A natural stopping point or a supplied stop sequence was reached. It is not a schema or factual-accuracy guarantee. | Parse the exact content, then check requested fields and values. Review any stop sequence you configured. |
content_filter | Content was omitted due to filtering. | Handle the provider's filtering outcome separately. Increasing the token limit is not a fix for this reason. |
tool_calls or legacy function_call | The assistant requested a tool step. | Handle the tool flow in your client. Tool arguments are a separate payload, not necessarily the final answer. |
| Missing or unfamiliar | The stopping condition is unconfirmed by this checker. | Read the provider's documentation and complete response. Do not infer truncation solely from token counts. |
The checker inspects only choices[0]. A single streaming chunk may contain a delta and no complete message; assemble the answer in your client or capture a non-streaming response before checking it.
Separate JSON syntax from missing fields
Parse the exact string in choices[0].message.content. Markdown fences, a sentence before the object, an unfinished string or a missing comma can prevent parsing. The checker flags these conditions without silently stripping text or inventing closing braces.
If JSON parses but your required field is absent, the syntax passed and the field checklist failed. That alone does not establish why the model omitted it. A field present with null is different from a missing key: null may correctly represent information that the source never supplied.
For example, {"name":"Demo"} is valid JSON. If your required checklist is name, count, it still lacks count. This is a handwritten illustration, not a live model result.
For DeepSeek JSON mode, check its own instructions
DeepSeek's JSON Output guide calls for response_format with type: json_object, explicit JSON instructions and an example in the prompt, and enough max_tokens for the intended output. It also documents possible empty content. JSON mode does not replace checking the actual response and your business fields.
Do not copy an output-token parameter blindly between every provider or model. Use the field and limit documented for the route you are actually calling.
Keep the repair smaller than the task
- Ask for fewer fields or fewer records when the permitted output is small.
- Separate a long document into bounded tasks only when you can preserve identifiers and source context.
- Handle missing, filtered and tool responses explicitly in your application.
- Review the source values after parsing. Valid JSON can still contain an incorrect price or invented fact.
The ChinaGenAI site route currently caps requested output at 256 tokens when enabled. A long extraction may not fit; key issuance can also be paused independently. Use API access for current availability and limits.
When the response is complete and your task is supplier-quote extraction, use the separate quote field checker for its types, numeric bounds and MOQ rules. For a first request rather than a failed result, follow the DeepSeek quote-to-JSON setup.
Official field definitions and JSON-mode instructions checked 19 October 2026. This page does not claim a measured failure frequency, model ranking or guaranteed search traffic.
Free API access and limits
Check the site free API access and limits before trying to obtain a site key. Availability follows the operator’s enablement switch and the published request limits. It is not a source of official provider keys.
ChinaGenAI is not affiliated with the providers named here. This release contains setup guidance and explicitly labeled fictional examples, not claimed live model benchmarks.