xAI releases Grok 4.5, GLM-5.2 trails on SWE Bench Pro

xAI has released Grok 4.5, a closed-weight model trained on tens of thousands of GB300s alongside Cursor. The launch highlights token efficiency as a key feature, claiming approximately 16k output tokens per task compared to Opus's 67k.

• Pricing is set at $2/M input and $6/M output, with availability in the EU delayed until mid-July.
• On SWE Bench Pro, Grok 4.5 scored 64.7%, trailing Fable (80.4%) and Opus 4.8 (69.2%), while GLM-5.2 scored 62.1%.
• The article notes that GLM-5.2 appears in xAI's own charts, positioned 2.6 points behind Grok 4.5 but ahead of GPT 5.5.
• Discrepancies exist between provider-harnessed scores and independent runs on DeepSWE benchmarks.

The release positions Grok 4.5 as a high-efficiency frontier model, though the source highlights significant performance gaps with competitors like Fable and questions regarding benchmark consistency.