xAI Launches Grok 4.5 and Prepares Versions 4.6 and 4.7 to Compete with Anthropic

1.5-trillion-parameter model bets on agentic coding and cost 3x lower than Claude Opus 4.8

PrezenceAI Editorial · Published on 13/08/2026

🤖

This article was produced by PrezenceAI with the assistance of artificial intelligence. The main information was extracted from sources cited at the end of this text. The analysis, synthesis, and writing are the sole editorial responsibility of PrezenceAI.

Quick read...

• xAI launched Grok 4.5 on July 8, 2026 as an efficient alternative for coding and agent tasks.• The model utilizes V9 Mixture-of-Experts architecture with 1.5 billion parameters and was trained with real data from Cursor.• Grok 4.5 has a context window of 500 thousand tokens and a configurable reasoning_effort parameter per call.• The Grok 4.5 API costs $2 per million input tokens and $6 per million output tokens.• Grok 4.5 achieved 54-56 on the Intelligence Index, occupying the 4th position among 185 models with leadership in agentic tool use.

🎧 Listen to this article

0:00 / --:--

SpaceXAI launched Grok 4.5 on July 8, 2026, positioning the model as a lower-cost, higher-efficiency alternative to top-tier models from Anthropic and OpenAI in coding and agentic tasks.

The model is based on the V9 Mixture-of-Experts architecture with 1.5 trillion parameters, co-trained with Cursor — a code development company SpaceXAI is acquiring for $60 billion — using real developer session data and agentic interactions. Training ran on tens of thousands of NVIDIA GB300 GPUs on the Colossus supercomputer. The context window is 500,000 tokens, with a configurable reasoning_effort parameter per call (low / medium / high).

API pricing was set at $2 per million input tokens and $6 per million output tokens — compared with $5/$25 for Claude Opus 4.8 and $30 per million output for GPT-5.6 Sol. xAI also claims 2x better token efficiency over direct competitors: the same task would consume half the tokens compared to rival models.

On independent Artificial Analysis benchmarks, Grok 4.5 scored 54-56 on the Intelligence Index, ranking 4th among 185 models at launch, with the top spot in agentic tool use. On SWE Marathon it posted a 29.0% resolution rate, above Claude Opus 4.8 (26.0%) and Claude Fable 5 (24.0%). On Terminal-Bench 2.1 it reached 83.3%. On SWE-Bench Pro and DeepSWE 1.1 it trailed Claude Fable 5 and GPT-5.6 Sol. Elon Musk described the model as equivalent to Opus 4.7 but faster.

Grok 4.5 was not available in the European Union at launch. The announced monthly cadence was confirmed: SpaceXAI launched Grok 4.6 on August 12, 2026, matching GPT-5.6 Sol on the Artificial Analysis index with 61 points.