llama-3.3-70b-versatile vs ministral-8b-latest

Capability for the money

Not the same tier: llama-3.3-70b-versatile leads ministral-8b-latest by 18.0 points on following instructions (IFBench), for 4.3× the price. They are level on coding (LiveCodeBench), tool use (τ²-bench) and hard reasoning (GPQA Diamond), so the premium only pays off on following instructions work.

1. Following instructions

   extraction, classification, sticking to a format

   IFBench
   llama-3.3-70b-versatile47.1%

   ministral-8b-latest29.1%

   llama-3.3-70b-versatile +18.0

2. Coding

   writing and fixing code

   LiveCodeBench
   llama-3.3-70b-versatile28.8%

   ministral-8b-latest30.3%

   ministral-8b-latest +1.5

3. Tool use

   multi-step agent work against real tools

   τ²-bench
   llama-3.3-70b-versatile26.6%

   ministral-8b-latest26.6%

   level

4. Hard reasoning

   graduate-level science and multi-step analysis

   GPQA Diamond
   llama-3.3-70b-versatile49.8%

   ministral-8b-latest47.1%

   llama-3.3-70b-versatile +2.7

Averaged over the benchmarks both models sat, ministral-8b-latest returns 221.8 points of pass rate per dollar of blended price against 59.5 — the ratio that decides whether the dearer model earns its rate.

Each figure is a pass rate: the share of that test set the model answered correctly, measured by a third party rather than self-reported by the lab. Rows are different tests of different difficulty, so compare a model against the other model, not one row against the next. And a benchmark is not your workload — it narrows the shortlist, it does not pick for you.