Grok 4.5 costs 80% less than Fable 5. Do benchmarks matter?

Key Takeaways

• Grok 4.5 costs $2/$6 per million tokens vs Fable 5's $10/$50, an 80% discount
• On DeepSWE 1.1, Grok 4.5 scores 53% vs Fable 5's 70%, a 17-point gap
• xAI claims Grok 4.5 uses 4.2x fewer tokens than Opus 4.8 on SWE Bench Pro tasks

xAI released Grok 4.5 on July 8, trained on tens of thousands of Nvidia GB300 GPUs. The model trails Anthropic's Fable 5 and OpenAI's GPT 5.5 on most coding benchmarks. But it costs a fraction of the price, and that gap may matter more than raw performance for teams running large-scale inference.

How does Grok 4.5 perform on coding benchmarks?

The results are mixed. On Terminal Bench 2.1, which tests complex command-line tasks, Grok 4.5 scores 83.3%. That nearly matches GPT 5.5 at 83.4% and trails Fable 5 at 84.3% by just one point. For CLI-heavy workflows, the three models are functionally equivalent.

The gaps widen on harder tests. DeepSWE 1.1 measures how well a model resolves real GitHub issues. Grok 4.5 hits 53%, compared to GPT 5.5 at 67% and Fable 5 at 70%. That's a 17-point deficit against the leader.

SWE Bench Pro, a curated set of difficult software engineering problems, puts Grok 4.5 at 64.7%. It beats Opus 4.8's 69.2% in some configurations but falls well short of Fable 5's 80.4%.

Why the pricing difference is so dramatic

Grok 4.5 costs $2 per million input tokens and $6 per million output tokens. Compare that to the competition. Opus 4.8 runs $5 input and $25 output. GPT 5.5 and GPT 5.6 sit at $5 input and $30 output. Fable 5 charges $10 input and $50 output per million tokens.

On output tokens alone, Grok 4.5 is 88% cheaper than Fable 5. That's not a rounding error. It changes the math on what tasks you can afford to automate.

Model: Grok 4.5, Input (per 1M tokens): $2, Output (per 1M tokens): $6. Model: Opus 4.8, Input (per 1M tokens): $5, Output (per 1M tokens): $25. Model: GPT 5.5 / GPT 5.6, Input (per 1M tokens): $5, Output (per 1M tokens): $30. Model: Fable 5, Input (per 1M tokens): $10, Output (per 1M tokens): $50.

xAI also claims Grok 4.5 uses 4.2 times fewer tokens than Opus 4.8 on SWE Bench Pro tasks and delivers results at 80 tokens per second. Lower per-token pricing combined with fewer tokens per task makes the effective cost difference even larger. If those efficiency claims hold in production, Grok 4.5 could cost 10-15x less than Fable 5 for equivalent workloads.

What tradeoffs should AI teams expect?

The pricing strategy echoes what Chinese vendors like Zhipu and DeepSeek have done: get close enough on performance, then win on price. It's a bet that most production use cases don't need the last 15% of benchmark performance.

That bet makes sense for certain workloads. If you're running agentic loops that make hundreds of API calls per task, cost dominates. A model that's 20% worse but 80% cheaper can be the right choice. But if you're building a product where a single failure mode causes user churn, the benchmark gap matters.

xAI says it relied on heavy data filtering, deduplication, and domain-specific selection during training to keep data quality high. The reinforcement learning stage covered hundreds of thousands of tasks, mostly from software engineering, with automated scoring. xAI built the infrastructure for asynchronous learning so agentic runs could stretch over many hours while training continued in parallel.

Where is Grok 4.5 available?

Grok 4.5 is available now through Grok Build, Cursor, and the xAI console. Plugins are live for Word, PowerPoint, and Excel. The model isn't available in the EU yet. xAI is targeting a mid-July launch for European users.

One notable detail: xAI trained Grok 4.5 alongside the code editor Cursor, which SpaceX acquired in mid-June for $60 billion in stock. That acquisition suggests xAI sees developer tooling as central to its strategy, not just model APIs.

Frequently Asked Questions

Is Grok 4.5 better than GPT 5.5?

On Terminal Bench 2.1, they're nearly identical (83.3% vs 83.4%). On DeepSWE 1.1, GPT 5.5 leads by 14 points. Grok 4.5 costs 80% less, so the answer depends on your workload and budget.

How much does Grok 4.5 cost compared to Fable 5?

Grok 4.5 charges $2 input and $6 output per million tokens. Fable 5 charges $10 input and $50 output. That's an 80-88% price reduction depending on your input/output ratio.

When will Grok 4.5 be available in Europe?

xAI is targeting mid-July 2026 for EU availability. The model launched on July 8 in other regions.

What hardware was Grok 4.5 trained on?

xAI trained Grok 4.5 on tens of thousands of Nvidia GB300 GPUs using asynchronous learning infrastructure.

Source: The Decoder / Matthias Bastian