Grok 4.5 is so cheap compared to Fable 5 and GPT 5.5 that benchmark gaps may not matter much

2026-07-08 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, quick

Summary

xAI has released Grok 4.5, a new large language model trained on tens of thousands of Nvidia GB300 GPUs, specifically targeting coding, agentic tasks, and knowledge work. While its benchmark performance presents a mixed picture, it nearly matches GPT 5.5 on Terminal Bench 2.1 with 83.3% but trails significantly on DeepSWE 1.1 (53% vs. GPT 5.5's 67% and Fable 5's 70%) and SWE Bench Pro (64.7% vs. Fable 5's 80.4%). Despite these performance gaps, Grok 4.5 is positioned as a highly cost-effective alternative, priced at \$2 per million input tokens and \$6 per million output tokens. This is substantially cheaper than competitors like Opus 4.8 (\$5 input, \$25 output) and Fable 5 (\$10 input, \$50 output). xAI also claims Grok 4.5 uses 4.2 times fewer tokens than Opus 4.8 on SWE Bench Pro tasks, delivering results at 80 tokens per second. The model is available via Grok Build, Cursor, and the xAI console, with EU availability planned for mid-July.

Key takeaway

For AI Engineers or ML Directors evaluating LLMs for coding and agentic workflows, Grok 4.5 presents a compelling cost-performance proposition. If your projects are budget-constrained or require high-volume token usage, you should consider Grok 4.5 despite its minor benchmark gaps. Its significantly lower per-token pricing and claimed token efficiency could lead to substantial operational savings, making it a viable alternative to more expensive models like Fable 5 or GPT 5.5.

Key insights

Grok 4.5 offers competitive performance at a significantly lower cost, challenging higher-priced LLMs.

Principles

• Price can offset minor benchmark deficits.
• Data quality and RL are key training steps.
• Asynchronous training supports agentic tasks.

Method

xAI used heavy data filtering, deduplication, and domain-specific selection, followed by reinforcement learning on hundreds of thousands of software engineering tasks with automated scoring. Asynchronous learning infrastructure supported long agentic runs.

In practice

• Evaluate Grok 4.5 for cost-sensitive coding tasks.
• Utilize Grok Build or Cursor for integration.
• Explore plugins for Word, PowerPoint, and Excel.

Topics

• Grok 4.5
• Large Language Models
• LLM Benchmarks
• AI Model Pricing
• Software Engineering AI
• Agentic AI

Best for: CTO, VP of Engineering/Data, MLOps Engineer, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Codex is gaining steam — AI platforms are evolving to be more accessible to non-technical users and offer competitive performance-to-cost ratios.
• GLM 5.2 Z.ai Coding Model. From Beginner to Power User. Locally & API & Benchmark — GLM 5.2 offers a cost-effective, high-performance coding model comparable to leading LLMs like Fable 5.
• GPT 5.5 Arrives, DeepSeek V4 Drops, and the Compute War Intensifies — Domain-specific training and cost-efficiency are becoming critical differentiators for new large language models amidst compute scarcity.
• Great news for xAI: Grok is now pretty good at answering questions about Baldur’s Gate — xAI's Grok excels in video game knowledge, reflecting a specific development focus despite leadership challenges.
• Grok 4.2: Powerful and Fast Coding Model! Most Powerful & Intelligent Model We've Seen? (Early Test) — Grok 4.2 is an upcoming model with enhanced multimodal capabilities and efficiency, despite some early coding limitations.
• Fable is Mythos, and it is really good. — Fable 5 sets a new benchmark for coding LLMs, offering advanced capabilities despite cost and data retention implications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.