orchid01
orchid01 is Orchid’s finance-native model. Use model id orchid01 in all requests.
There are two modes — standard and thinking — controlled via the orchid config object in the request body.
Standard mode (default)
Fast, direct responses. Recommended for most tasks: document extraction, summarisation, Q&A over provided context, structured data conversion.
response = client.chat.completions.create(
model="orchid01",
messages=[{"role": "user", "content": "Extract all covenants from this agreement..."}],
# thinking is off by default
)
| Setting | Value |
|---|
Default max_tokens | 4096 |
Default temperature | 0.1 |
Recommended max_tokens | 4096–8192 |
Thinking mode
Extended reasoning for complex multi-step analysis. The model reasons through the problem before responding. Reasoning is visible in reasoning_content alongside the final answer in content.
response = client.chat.completions.create(
model="orchid01",
messages=[{"role": "user", "content": "Build a full covenant analysis..."}],
extra_body={"orchid": {"thinking": True}},
)
# Access reasoning and answer separately
reasoning = response.choices[0].message.model_extra.get("reasoning_content", "")
answer = response.choices[0].message.content
Thinking mode requires max_tokens ≥ 16,000. Reasoning tokens count toward the limit — responses may be truncated below this threshold.
| Setting | Value |
|---|
Minimum max_tokens | 16000 |
Recommended max_tokens | 16384–32768 |
temperature | Fixed at 1.0 (set automatically) |
If you pass max_tokens below 16,000 with thinking enabled, Orchid automatically raises it to 16,000 and includes max_tokens_adjusted: true in the response metadata.
The orchid config object
Pass an orchid key alongside standard OpenAI fields to control Orchid-specific behaviour:
{
"model": "orchid01",
"messages": [...],
"orchid": {
"thinking": false,
"dehallucinate": true
}
}
| Field | Type | Default | Description |
|---|
thinking | boolean | false | Enable deep reasoning mode |
dehallucinate | boolean | true | Run grounding check and return hallucination_score |
orchid field with grounding information:
{
"choices": [...],
"orchid": {
"dehallucinate_requested": true,
"grounded": true,
"score": 0.97,
"flagged_spans": [],
"checked": true
}
}
| Field | Description |
|---|
grounded | Whether all claims are supported by provided context |
score | Confidence score 0.0–1.0 (1.0 = fully grounded) |
flagged_spans | Specific claims that could not be verified |
checked | Whether the grounding check actually ran |
dehallucinate: true (default), the response contains financial figures, and sufficient context was provided. If skipped, checked: false and score: 1.0.
