Claude Sonnet 5 – benchmark results

https://artificialanalysis.ai/models/claude-sonnet-5

32•lucamark•1h ago

Comments

iLoveOncall•1h ago

Half of the data is missing and the rest is inconsistent between different graphs and sections. Is the benchmark having Sonnet 5 generate the page and seeing how many hallucinations it has?

Tiberium•1h ago

Seems like the model is incredibly inefficient at max reasoning, and even at high/xhigh it uses far more tokens than other models, including Gemini 3.5 Flash, GLM 5.2 and so on. GPT 5.5's efficiency in tokens is still unmatched.

trentor•1h ago

Same with opus nothing above medium has a reasonable improvement for the tokens spent.

atemerev•1h ago

Yet another mediocre model. Mostly irrelevant among open weights alternatives. Fable wen.

butterisgood•43m ago

I used sonnet five today to evaluate work I’m doing on an experimental programming language with an interesting concurrency model.

I asked it to try to figure out why one of the examples wasn’t working.

It read the implementation of the compiler and the runtime, found the bug, fixed it, fixed the example and the only thing I had to do manually is suggest a less silly name for a particular function.

I would use sonnet 5 for coding … seems alright!

lucamark•40m ago

Agree. It is a mediocre model, expensive while not being a frontier

datakan•1h ago

I'm so sick of Anthropics usage caps and how their model devours tokens.

system2•56m ago

It starts with NVIDIA artificially and slowly releasing its tech. If the GPUs were cheaper, we would have better models by many other companies, and competition would take care of these greedy tactics.

lucamark•36m ago

Remember that such models are available to us thanks to NVIDIA GPUs

trentor•25m ago

Technology doesn't pause for 50 years just because you remove one key player. After their antics during crypto and now during AI I wished we had a different hardware provider at the helm.

CSMastermind•57m ago

Using Fable, pretty much every request hit some gate they had for no discernible reason. These provider-level rejections should be incorporated into benchmarks as 0s on the tasks since that's the experience you'll actually get using the model.

cjk•40m ago

I have heard this from a bunch of folks, but that was not my experience. For the couple days I was able to use it, I didn't hit a single gate, and I was using it pretty extensively (but not for anything security-related).

lucamark•37m ago

Never had rejections in the short time Fable was available

nsingh2•55m ago

Cost per task is shockingly high. More expensive than Opus 4.8, second in place to Fable.

Cost per task data is only available for max effort though, might just be very inefficient at that effort level.

DrProtic•51m ago

I feel like they repackaged Opus, slightly nerfed it, and reduced price per token.

A release just to have a headline while Fable situation is getting resolved.

Claude Sonnet 5

Claude Code is steganographically marking requests

Meta's brain-scanning system reads sentences non-invasively, code open source

Claude Science

Nano Banana 2 Lite

How does a pull-back car work? Illustrated teardown

I ported Kubernetes to the browser

I built a mmWave material classification radar (2025)

Stroustrup's Rule (2024)

Building a custom octocopter from scratch with no prior hardware experience

Show HN: My 13-year-old built an ant colony tracker

Knoppix

Long Island's decommissioned nuclear power plant

CERN bids farewell to the LHC and enters Long Shutdown 3

Waveloop: What Fable left me

Memoirs of Extraordinary Popular Delusions and the Madness of Crowds (1852)

Set up your own DoH (DNS over HTTPS) service

Have you restarted your computer this week?

Reading the internals of Postgres: Database cluster, databases, and tables

Matrix URIs, a URL syntax from Tim Berners-Lee that never shipped (1996)

I built a 10 inch mini rack from aluminium extrusions

Morbid: Debunking Modern Longevity Science

A peek into Reddit's anti-spam internals

Amazon seller reveals glimpse of shadow bribery market

RF hacking my cloud-controlled ceiling fan

1.38 Millimeter Microcontroller

Zluda 6 release (run unmodified CUDA applications on non-Nvidia GPUs)

Tokyo has only two barley tea makers, we visited one to see how mugicha is made

SedonaDB 0.4: GPU-accelerated spatial joins

A Fake Shell for Pangenomics