Comment by nearbuy

I do not (generally) trust benchmarks. I only trust what a model does with MY code.

Forget DS. I asked MiMo 2.6 yesterday to explain ML/LLMs to me succinctly and the pointed it at Karpathy's micrograd code. It produced a C implementation called `xor_mlp`, a tiny model that learnt how `xor` worked. I then asked it to produce a model that can play tictactoe without losing (mostly). It did. It supervised the training process and produced a compiled version with multiple switches. The pi-dev session is still running, so here are actual stats

↑45k ↓35k R1.0M CH99.4% $0.019 4.2%/1.0M (auto) - (opencode-go) mimo-v2.6-flash • high

And here is Luna on the same workflow (I had to poke and prod a bit to get what I wanted):

↑141 ↓34k R1.0M W43k CH95.3% $0.072 4.2%/1.1M (auto) (opencode-go) gpt-5.6-luna • high

I expect similar results from DS41F/MS13. Closer to MiMo costs than Luna.

So the "significantly cheaper" thing may not really hold, more so when Luna has to actually read my codebase to do the stuff that I want rather than rely on world knowledge. The 8-10x cache read cost differential itself will kill the token budget.

Artificialanalysis benchmark is a combination of a several benchmarks which might or might not represent realistic coding:

"Artificial Analysis Intelligence Index combines performance across 10 evaluations: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, and AA-LCR v1.1."

Not saying it doesnt have any value but it's probably irrelevant if you use these AIs for a specific use case. Like for example Humanity Last Exam tests general knowledge, which is not very useful for coding.

It's best to go to the specific coding benchmarks and compare there.

Does it use less tokens or we just get no accounting of the thinking tokens in OpenAI/Claude models?