SpaceXAI releases Grok 4.5 with improved coding performance and lower costs

1. Performance and Market Positioning

SpaceXAI’s Grok 4.5 has secured the fourth position on the Artificial Analysis Intelligence Index, trailing only Fable 5, GPT-5.5, and Opus 4.8. The model demonstrates significant improvements over its predecessor, Grok 4.3, gaining 16 points on the index. It currently outperforms all open-weights models and Google’s Gemini series, establishing SpaceXAI as a key competitor in the intelligence frontier. Grok 4.5 features 1.5 trillion parameters, making it three times larger than the previous version, though its context window has been reduced from 1 million to 500,000 tokens.

2. Coding and Agentic Capabilities

Grok 4.5 shows strong performance in agentic tasks, particularly in knowledge work and coding. When utilized within the Grok Build harness, the model ranks on par with GPT-5.5 in Codex for coding tasks, scoring 76 on the Artificial Analysis Coding Agent Index. It also achieved the top score of 33% on the 𝜏³-Banking benchmark. While the model shows increased accuracy in knowledge-based tasks, its hallucination rate rose from 25% to 54% compared to Grok 4.3, a trend often observed in larger, more confident models.

3. Cost Efficiency and Token Usage

A primary advantage of Grok 4.5 is its cost-effectiveness. The model is priced at $2 per 1 million input tokens and $6 per 1 million output tokens, with a 75% discount for cache hits. It maintains a headline price over 60% lower than Claude Opus 4.8 and GPT-5.5. In coding applications, Grok 4.5 is notably efficient, costing $2.49 per task compared to $5.07 for GPT-5.5 and $11.80 for Fable 5. This efficiency is attributed to both lower pricing and a significantly reduced token usage rate, as the model required only 1.9 million tokens per task in coding evaluations, compared to 6.2 million for GPT-5.5 and 7.2 million for Fable 5.