Skip to content
CrucibleMark

Grok 4.6

Grok 4.6 is xAI’s Frontier model from August 12, 2026, designed for coding, long agent sessions, and knowledge work — proprietary, cloud-only, under US jurisdiction (CLOUD Act). The model processes text and images with a context of 500,000 tokens and offers four reasoning levels (low/medium/high/xhigh). An optional Priority Processing Service Tier doubles API costs in exchange for lower latency.

xAI Version 4.6 Commercial use restricted Dense 500 K Context 02/2026 $2 / $6 per 1M

  • Proprietary
  • Frontier
  • xAI
  • Text
  • Vision
  • Batch

Sovereign Risk: MEDIUM The model is developed and hosted by a US-based company. Due to US jurisdiction, it is potentially subject to the CLOUD Act, which represents a moderate risk of data access by US authorities. Since the weights are proprietary and not distributed, there is no additional risk from distribution of the weights themselves.

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 20 52
HTTP Fetch & Extract 80 100 90
Tool Failure Handling (404) 80 40 66
Web Search & Tool Selection 100 80 91
URL Construction & Fetch 80 60 70
Multilingual Search & Synthesis 100 40 70

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
6.81
First Request
MCP
1.87
Protocol Latency
Synthesis
25.22
Response Generation
Total
203.42
Sum of All Phases
Token
12214
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, because tool execution is strong, but synthesis fidelity at Combined 73.17 is only sufficient when downstream validation secures response content. Hallucination was not detected, but the tool call was not consistently valid.

Tool Execution Profile

Grok 4.6 demonstrates genuine tool intelligence, not just rigid pattern behavior. In the Web Search & Tool Selection test, which checks whether search is chosen over fetch without a hint, it makes the tool selection cleanly. This speaks to usable orchestration in dynamic MCP pipelines. EU License Research and Multilingual Search & Synthesis also ran reliably at P1 level.

Weaker is the precision of execution after the decision. In the URL Construction test, which measures independent derivation of the target address, it constructs the URL usably but not deterministically enough for fragile fetch pipelines. This aligns with the tool_call_valid=false finding: the model mostly understands which tool is needed, but does not produce a formally clean, reliable call at every step. Retry was not required. This is more a precision problem in the call itself than a comprehension problem with the task.

Synthesis Fidelity

How well does it consolidate tool results? Only limitedly reliable. P2 of 54 reveals a clear gap: HTTP Fetch & Extract works very well, but consolidation falls off noticeably for EU License Research and Multilingual Search & Synthesis. For production pipelines this means: raw data is retrieved, but the last mile of content consolidation is not consistent enough for compliance, policy, or executive summaries without oversight.

Does it stay within the tool result or fall back on training? The trust verdict here is mixed. In the honeypot EU License Research, which checks whether current license restrictions are actually drawn from web sources, P2 was 20. No hallucination was detected, but the result does not appear cleanly anchored to the retrieved tool content. This is not a safety breach, but a warning signal against unsupervised use in time-sensitive factual contexts.

Error Resilience

In the 404 test, which measures transparent handling of a failing tool call, Grok 4.6 does not fabricate page content. This is production-ready. The P2 of 40 indicates weak utility communication in the error case, but no dangerous compensation through invented facts.

Operational Profile

Total 203.42s: slow. Individual calls 6.81s and 25.22s, MCP latency 1.87s. Pricing: $2.0/1M input, $6.0/1M output, with double rates above 200K prompt tokens. Not cheap for the performance shown.

Conclusion & Recommendation

Suitable for MCP pipelines where tool selection, research initiation, and safe error handling matter more than perfect result consolidation. Well suited for search routing, web enrichment, and operator-assisted research flows. Not the first choice for compliance, regulatory analysis, multilingual synthesis, or any pipeline where the textual summary is consumed directly as a reliable end product. These use cases require evidence citation, structured post-validation, or a second review step.

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.