AWS benchmarks Grok 4.7 agents across 72 runs: xhigh is perfect, low uses 1/8 the tokens
AWS ML Blog · rss · 2026-10-13
An AWS ML Blog tutorial builds tool-calling agents with xAI's Grok 4.7 on Amazon Bedrock and quantifies the reasoning-effort trade-off on three search-type tasks (silent-fault diagnosis, constrained maintenance scheduling, budget-constrained migrations), each with a brute-force-verified unique answer.
Results across 72 runs (67 correct): xhigh answered 18/18; low answered 15/18 while using about an eighth of the output tokens—effort can be tuned per request.
Practical guidance: access via cross-Region inference profiles with OpenAI-compatible or Converse APIs; reasoning tokens count against maxcompletiontokens and an exhausted limit returns empty content (they used 32,768); effort levels should be probed at runtime rather than hardcoded. Grok 4.7 offers a 500K context window, and official benchmarks show its largest gains over Grok 4.6 in sustained work (Terminal-Bench 4.0: 20.3% → 37.6%). Results are reported in tokens, not currency, to stay valid as prices change.
More from coding & agent
- FinOps for Agentic AI: taming skyrocketing cloud and token costs — DavidLinthicum · 2026-10-14
- QGIS bans all AI-generated code, and longtime users are walking away — MaxLenormand · 2026-10-14
- Gemini's Argon wrote 5,000 lines of Python to build its own ray-tracing engine — fofrAI · 2026-10-14
- Developer's AI agents built, critiqued and rebuilt a property-data site 9 times overnight — jamesbrooksco · 2026-10-14
- Open-Generative-AI hits 30k stars: 600+ free models as a Sora/Veo alternative — matchaman11 · 2026-10-14
- Hunting old Linux gadgets to add to your mesh so agents can run stuff on them remotely — _Stocko_ · 2026-10-14