Skip to main content
gpt-5-2 sits between gpt-5-4-mini and gpt-5-4. It’s the cheapest GPT-5 variant with built-in web search and is OpenAI’s preferred model for short-to-medium agent loops that need to look something up before answering. If you don’t need browsing, gpt-5-4-mini is cheaper for the same quality on closed-book tasks. If you need flagship reasoning headroom, jump to gpt-5-4. Pricing: 1.75/1Minput,1.75 / 1M input, 14.00 / 1M output, $0.175 / 1M cache read — see the rate card.

​
Protocols

ProtocolPath
OpenAI Chat CompletionsPOST https://llm.bytespike.ai/v1/chat/completions
OpenAI ResponsesPOST https://llm.bytespike.ai/v1/responses

​
Quickstart

curl https://llm.bytespike.ai/v1/chat/completions \
  -H "Authorization: Bearer $BYTESPIKE_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "gpt-5-2",
    "messages": [
      { "role": "user", "content": "What is the current price of Bitcoin in USD?" }
    ],
    "tools": [{ "type": "web_search" }]
  }'

​
Capabilities

CapabilitySupported
Chat Completions✅
Responses API✅
Streaming (SSE)✅
Vision✅
Tool use (function calling)✅ parallel
JSON mode✅
Structured outputs (json_schema)✅
Reasoning effort✅ (low / medium / high)
Web search (built-in tool)✅
Context window128K tokens

​
When to use

  • Short agent loops with one or two browsing hops — pricing the web, looking up live data, fact-checking a draft. Cheaper than gpt-5-4 for the same kind of work.
  • Vision + lightweight reasoning — captioning, OCR, “what’s on this screenshot” questions where you don’t need flagship-level analysis.
  • Mid-volume RAG — when retrieval is doing the heavy lifting and the LLM is summarising / synthesising.
When not to use:
  • Hard reasoning that wants reasoning_effort: "high" continuously — pay the premium for gpt-5-4 or gpt-5-5.
  • High-volume classification / routing — gpt-5-4-mini is the right cost tier.

​
Next