Skip to main content
Pawa AI Reasoning capabilities helps the model plan before answering. When enabled, the model allocates extra compute for internal planning and verification, which typically improves accuracy on multi‑step tasks like analysis, transformation, and decision support.
Reasoning Example
Web Browsing Example Reasoning is the next step of CoT (Chain of Thought), naturally used to describe the logical steps generated by the model before reaching a conclusion. Reasoning strengthens this characteristic by going through training steps that encourage the model to generate chains of thought freely before producing the final answer. This allows models to explore the problem more profoundly and ultimately reach a better solution to the best of their ability by using extra compute time to generate more tokens and improve the answer—also described as Test Time Computation.

​
When to use Reasoning

  • Complex prompts that require multi‑step thinking (summarize → compare → decide). A good example is in Agentic AI
  • Data extraction with consistency checks and light validation
  • Tool‑using agents that must plan which tools to call and in what order
Reasoning adds latency and may consume more tokens. Use only when the task benefits from deeper planning

​
Reasoning Models in Pawa AI.

Currently we have one reasoning model:
  • pawa-v1-blaze-20250318: A powerful small language model (SLM) optimized for reasoning, complex generation, multimodal, tools understanding, agentic workflow, and advanced knowledge tasks.

​
Enable Reasoning

Add the optional capabilities field to your request. Set the name to reasoning and choose an effort level. Effort levels (guidelines):
  • low: small planning budget; minimal latency increase
  • medium (default): balanced quality vs. latency
  • high: larger budget for complex tasks; highest latency

​
Example Reasoning Request

{
  "model": "pawa-v1-blaze-20250318",
  "capabilities": {
    "name": "reasoning",
    "effort": "medium"
  },
  "messages": [
    {"role": "system", "content": "You are a careful analyst. Always verify numbers."},
    {"role": "user", "content": "Compare these quarterly figures and tell me the key trends: Q1: 124, Q2: 171, Q3: 163, Q4: 205."}
  ]
}

​
Examples Reasoning request in Pawa AI

curl --request POST \
  --url https://api.pawa-ai.com/v1/chat/request \
  --header "Authorization: Bearer $PAWA_AI_API_KEY" \
  --header 'Content-Type: application/json' \
  --data '{
  "model": "pawa-v1-blaze-20250318",
    "capabilities": { "name": "reasoning", "effort": "medium" },
    "messages": [
      {"role": "system", "content": "You are a careful analyst. Always verify numbers."},
      {"role": "user", "content": "Compare Q1=124, Q2=171, Q3=163, Q4=205 and summarize trends."}
    ]
  }'
Example assistant output (shortened): 
Growth accelerates through the year. Largest jump Q1→Q2 (+47). Q3 dips slightly (−8), then Q4 reaches a new high (+42 over Q3). Overall upward trend.

​
Prompting tips

  • Give concrete goals and constraints (“compare, then list 3 key trends”).
  • Ask for a structured output (bullets or JSON) to reduce ambiguity.
  • Combine with Tools Calling for grounded answers (e.g., fetch data, then analyze).

​
Performance & limits

  • Effort increases latency and token usage; prefer low for lightweight tasks.
  • Reasoning works with multimodal inputs and tool calls.
  • Streaming is supported; the final answer streams as usual.

​
Troubleshooting

  • Output feels generic: raise effort or tighten instructions with examples.
  • Too slow: lower effort or trim the context to essential inputs.
  • Inconsistent numbers: add a system rule to verify calculations and show working as bullet points (not raw scratchpad).

​
Example: Reasoning + Tools Calling

Reason over live data by enabling reasoning and offering a retrieval tool. 
curl --request POST \
  --url https://api.pawa-ai.com/v1/chat/request \
  --header "Authorization: Bearer $PAWA_AI_API_KEY" \
  --header 'Content-Type: application/json' \
  --data '{
  "model": "pawa-v1-blaze-20250318",
    "capabilities": { "name": "reasoning", "effort": "low" },
    "tools": [{ "type": "pawa_tool", "name": "web_search_tool" }],
    "messages": [
      {"role":"user","content":"Summarize today\'s top Pawa AI news in 3 bullets for a CTO."}
    ]
  }'
The model will plan, search, and produce a concise, CTO‑friendly summary.
⌘I