Code-specialist versus generalist small tier for coding workloads.
codestral-latest · mistral
gpt-5.4-mini · openai
Benchmark
| Workload Scenario | Codestral | GPT-5.4 mini | Price Delta |
|---|---|---|---|
| 1M input tokens (raw text) | $0.30 | $0.75 | −$0.45 |
| 1M output tokens (generation) | $0.90 | $4.50 | −$3.60 |
| 1M tokens · 70% input / 30% output mix | $0.48 | $1.875 | −$1.395 |
| Standard chat turn (4K in / 800 out, 50% cached) | $0.00138 | $0.00525 | −$0.00387 |
| Monthly scale (10K requests / day) | $414.00 | $1,575.00 | −$1,161.00 |
Negative difference = Codestral is cheaper. Positive = GPT-5.4 mini is cheaper.
Scaling Curve
Total cost of a token volume at 70/30 input/output split (uncached). Log-log scale.
Verdict
Codestral is cheaper on both input and output rates, so it costs less at every input/output mix. Price alone still isn't the whole decision: capability, latency and context limits (Codestral: 256K, GPT-5.4 mini: 1.1M) may justify the premium for your task.
FAQ
On input, Codestral is cheaper ($0.30/M vs $0.75/M). On output, Codestral is cheaper ($0.90/M vs $4.50/M). The same model is cheaper on both sides, so it wins at every mix.
A chat-style request (4,000 input + 800 output tokens, 50% cached) costs $0.00138 on Codestral and $0.00525 on GPT-5.4 mini — Codestral is 3.8× cheaper for that workload.
Codestral supports 256,000 tokens (32.8K max output); GPT-5.4 mini supports 1,050,000 (128K max output). GPT-5.4 mini fits 4.1× more context, which matters for long documents and agents.