cd /news/large-language-models/grok-4-5-didn-t-win-a-single-benchma… · home topics large-language-models article
[ARTICLE · art-53741] src=pub.towardsai.net ↗ pub= topic=large-language-models verified=true sentiment=· neutral

Grok 4.5 Didn't Win a Single Benchmark — Then Undercut Opus by 17x on Cost Per Task

SpaceXAI released Grok 4.5, which placed fourth on the Artificial Analysis Intelligence Index behind Fable 5, GPT-5.5, and Opus 4.8, but undercut Opus by 17x on cost per task for coding tasks. The model's cost efficiency, not benchmark scores, positions it as a significant release for API users.

read1 min views1 publishedJul 10, 2026
Grok 4.5 Didn't Win a Single Benchmark — Then Undercut Opus by 17x on Cost Per Task
Image: Pub (auto-discovered)

Member-only story

SpaceXAI’s new model finished behind Fable 5, GPT-5.5, and Opus 4.8 on the leaderboard. Then I ran the cost math and it resolved the same coding tasks for roughly 17x less than Opus. Here’s the number nobody is putting in the headline.

On Wednesday, Elon Musk’s SpaceXAI shipped Grok 4.5 — its first model since the company went public — and called it an “Opus-class model, but faster, more token-efficient and lower cost.” My first reaction was to roll my eyes. Every lab says its new model is “class-leading.” Then I looked at where Grok 4.5 actually lands on the leaderboards, and the honest answer is: fourth.

Not first. Not second. On the independent Artificial Analysis Intelligence Index it scored 54, placing fourth behind Fable 5, GPT-5.5, and Opus 4.8. On xAI’s own published coding benchmarks, Anthropic’s Fable 5 wins all four outright. Grok 4.5 doesn’t top a single chart.

And yet after I spent an afternoon with the numbers, I think it’s one of the most important releases of the year — because the metric that matters for anyone actually paying an API bill isn’t the benchmark score. It’s the cost to finish the task. And on that metric, Grok 4.5 quietly humiliates the models that beat it on paper.

The setup: a brutal week to launch a “good enough” model #

The timing is savage. Grok 4.5 landed the same week OpenAI was clearing GPT-5.6…

── more in #large-language-models 4 stories · sorted by recency
── more on @spacexai 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/grok-4-5-didn-t-win-…] indexed:0 read:1min 2026-07-10 ·