Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/phpaisdk/openai/llms.txt

Use this file to discover all available pages before exploring further.

OpenAI’s o-series models (o1, o3, o4-mini, and others) support a reasoning_effort parameter that controls how much compute the model spends on internal chain-of-thought reasoning before producing a response. More effort generally yields higher-quality answers at the cost of latency and tokens. The PHP AI SDK exposes this through the portable Reasoning::effort() helper, which maps cleanly onto OpenAI’s native reasoning_effort field.

​
Setting Reasoning Effort

Pass a Reasoning::effort() instance to the ->reasoning() method on your generation request. The SDK writes the value directly to reasoning_effort in the Chat Completions request body. 
use AiSdk\Generate;
use AiSdk\OpenAI;
use AiSdk\Reasoning;

$result = Generate::text('Solve this step by step: 17 * 23.')
    ->model(OpenAI::model('o4-mini'))
    ->reasoning(Reasoning::effort('medium'))
    ->run();

echo $result->text;

​
Effort Levels

OpenAI accepts three named effort levels. Choose based on your latency and quality requirements.
LevelDescription
lowFaster, less thorough — best for simple tasks where speed matters
mediumBalanced — a sensible default for most reasoning workloads
highSlower, most thorough — best for complex multi-step problems

​
Token Budget Limitation

The portable Reasoning::budget(int $tokens) form is not supported by the OpenAI provider. If you pass a token budget, the SDK throws an InvalidArgumentException before making any HTTP request:
OpenAI reasoning does not accept portable token budgets. Use Reasoning::effort(...) or provider raw options.
Use Reasoning::effort() instead, or pass OpenAI-specific reasoning controls via ->providerOptions('openai', ['raw' => [...]]).
The following code will throw an exception: 
use AiSdk\Generate;
use AiSdk\OpenAI;
use AiSdk\Reasoning;

// ❌ This throws InvalidArgumentException on the OpenAI provider.
Generate::text('Think briefly.')
    ->model(OpenAI::model('o3-mini'))
    ->reasoning(Reasoning::budget(4096))
    ->run();

​
Accessing Reasoning Tokens

After a successful call, the number of tokens consumed by the model’s internal reasoning is available on the usage object. The SDK reads this from completion_tokens_details.reasoning_tokens in the API response. 
use AiSdk\Generate;
use AiSdk\OpenAI;
use AiSdk\Reasoning;

$result = Generate::text('Prove that √2 is irrational.')
    ->model(OpenAI::model('o3'))
    ->reasoning(Reasoning::effort('high'))
    ->run();

echo $result->text;
echo "\n\nReasoning tokens used: " . $result->usage->reasoningTokens;
$result->usage->reasoningTokens is an integer when the field is present in the API response, or null when the model does not report it.

​
Compatible Models

The following model patterns registered in the OpenAI provider carry the reasoning capability. Any model whose ID matches one of these patterns supports Reasoning::effort().
Model PatternNotes
o*All o-series models — o1, o3, o4-mini, and future variants
gpt-5*GPT-5 family and all its variants
You can verify capability support at runtime before sending a request: 
use AiSdk\Capability;
use AiSdk\OpenAI;

$model = OpenAI::model('o4-mini');

if ($model->supports(Capability::Reasoning)) {
    // Safe to call ->reasoning(Reasoning::effort('medium'))
}
Reasoning streaming is supported — ReasoningDeltaPart chunks are emitted during a stream and contain the model’s internal reasoning text. Use ->stream() on your generation request to receive them incrementally.

Build docs developers (and LLMs) love

Get started for freeTalk to us