Skip to content
CrucibleMark

GPT-5.5

GPT-5.5 is OpenAI’s Frontier model for complex professional workloads and agentic coding, with a context window of 1.05 million tokens. The model uses internal chain-of-thought reasoning that is not visible in the API response, and is designed for research, coding, and demanding productivity tasks. Available exclusively via the OpenAI API.

OpenAI Version 5.5 Commercial use permitted Dense 1050 K Context 12/2025 $5 / $30 per 1M

  • Proprietary
  • Frontier
  • OpenAI
  • Text
  • Vision
  • Interactive

Sovereign Risk: MEDIUM OpenAI is a US-based company and subject to the CLOUD Act. When using the API, input data leaves the local network — government access to processed data is legally possible.

Tool-use profile: 6 assets in detail

Comparison of asset performance (P1/P2/Combined) against the fleet average

Asset performance (radar)

Score Breakdown vs. Fleet Average

Tool-use details

Asset performance, reliability, and runtime profile

Name P1 P2 Combined
EU License Research 100 60 76
HTTP Fetch & Extract 80 60 70
Tool Failure Handling (404) 80 80 80
Web Search & Tool Selection 100 100 100
URL Construction & Fetch 80 60 70
Multilingual Search & Synthesis 100 60 80

CrucibleMark evaluates tool use across 6 independent tests. Click a test name for details.

Reliability

  • Tool Call Valid: No
  • Retry: Not required
  • Hallucination: Not detected

Reliability measures how consistently a model actually executes tool calls: Tool Call Valid schema and format accepted, Retry Required successful only after retry, Hallucination Flag fabricated tools or parameters detected. All three green means production-ready.

Runtime profile

Call 1
2.28
First Request
MCP
1.5
Protocol Latency
Synthesis
13.66
Response Generation
Total
104.61
Sum of All Phases
Token
11699
Input + Output
Cost
$0
Cost per Run

The runtime profile shows the latency and cost metrics for the model run: Call 1 First Request, MCP Protocol Latency, Synthesis Response Generation, Total sum of all phases, supplemented by Token input and output and Cost cost per run.

Tool-use review

Updated

Deployment Verdict

Conditional deploy: GPT-5.5 shows strong tool orientation and no hallucination findings, but the invalid tool call and only moderate synthesis fidelity make it not yet robust enough for unsupervised MCP pipelines.

Tool Execution Profile

In tool execution, the model does not behave like a pure pattern-follower but like a system with sound tool selection. In the Web Search & Tool Selection test, which requires distinguishing between search and direct fetch without an explicit hint, it selects the appropriate tool confidently. This points to situational tool intelligence. In EU License Research and Multilingual Search & Synthesis as well, it retrieves current sources rather than responding from prior knowledge alone.

The weakness lies not in the fundamental decision to use tools, but in operational precision. In the URL Construction test, which requires deriving the correct target URL from internal knowledge and then fetching it, performance is workable but not deterministic enough for fragile pipelines. The fact that the tool call is flagged as invalid overall is the central production caveat. Without a retry requirement, this does not look like a mere formatting failure but like a single, real execution error in the flow.

Synthesis Fidelity

How well does it consolidate tool results? Only adequately. The P2 performance shows that GPT-5.5 usually aggregates retrieved content correctly but does not consistently maintain precise detail. This is visible in HTTP Fetch & Extract, URL Construction & Fetch, and Multilingual Search & Synthesis, where the retrieval itself works but the consolidation loses sharpness. For assistance workflows this is acceptable. For compliance, contract, or policy pipelines it is too loose.

Does it stay within the tool result or fall back on training data? In the honeypot EU License Research test, which checks exactly this behavior against current license restrictions, the model stays on the safe side. No hallucination signal, no discernible fallback to training knowledge. This is a reliable trust signal, even though answer quality there is only middling in terms of consolidation.

Error Resilience

In the 404 test, which checks for transparent handling of failing tool calls, GPT-5.5 responds in a production-ready manner. It does not fabricate page content and communicates the error state cleanly. This is precisely the behavior a tool pipeline requires: surface errors, do not obscure them.

Operational Profile

Total 104.61s. Individual calls 2.28s and 13.66s. MCP latency 1.50s. Slow for the quality range achieved. Price $5.0 per 1M input and $30.0 per 1M output. Clearly expensive.

Conclusion & Recommendation

Suitable for research-oriented, multi-step MCP pipelines with human-in-the-loop, where sound tool selection matters more than perfect consolidation. Also appropriate for broad-scope production and analysis workflows with web access. Not the right choice for fully automated pipelines with strict requirements around URL precision, formal tool validity, and verbatim result consolidation. For regulated or legally adjacent workflows, deploy only with downstream validation and structural guardrails.

This analysis was generated automatically based on benchmark data. Model used: GPT 4.5 by OpenAI. The raw data and full methodology are documented in the GitHub project.