Claude Haiku 4.5
VS
Mistral Medium 3.1
Claude Haiku 4.5 vs Mistral Medium 3.1: which wins at real work?
22 task areas · same graded test runs · rank comparison only, so 0–100 and Elo collections never mix raw scores.
Claude Haiku 4.5 wins 21 of 22 task areas we tested; Mistral Medium 3.1 takes 1. Mistral Medium 3.1 costs 2.5× less per token ($2.4 vs $6 per 1M).
Mistral Medium 3.1 costs 2.5× less per token ($2.4 vs $6 per 1M).
Task by task
| Task area | Claude Haiku 4.5 | Mistral Medium 3.1 | Winner |
|---|---|---|---|
| Research & Competitive Analysis |
#59
/ 110
Usable
|
#105
/ 110
Weak
|
Claude Haiku 4.5 |
| Data & Analytics |
#53
/ 110
Excellent
|
#87
/ 110
Excellent
|
Claude Haiku 4.5 |
| AI Strategy |
#91
/ 126
Strong
|
#120
/ 126
Usable
|
Claude Haiku 4.5 |
| Frontend & Landing Pages |
#7
/ 109
Usable
|
#35
/ 109
Needs editing
|
Claude Haiku 4.5 |
| Translation & Localization |
#27
/ 110
Excellent
|
#53
/ 110
Excellent
|
Claude Haiku 4.5 |
| Legal & HR |
#48
/ 110
Excellent
|
#70
/ 110
Strong
|
Claude Haiku 4.5 |
| Content & Brand |
#89
/ 124
Usable
|
#110
/ 124
Needs editing
|
Claude Haiku 4.5 |
| Executive Assistant |
#73
/ 112
Strong
|
#94
/ 112
Usable
|
Claude Haiku 4.5 |
| Structured Output |
#109
/ 113
Usable
|
#91
/ 113
Strong
|
Mistral Medium 3.1 |
| Presentations & Decks |
#88
/ 110
Strong
|
#105
/ 110
Usable
|
Claude Haiku 4.5 |
| Creative & Comedy | #84 / 110 | #100 / 110 | Claude Haiku 4.5 |
| Coding |
#94
/ 115
Usable
|
#108
/ 115
Needs editing
|
Claude Haiku 4.5 |
| RAG, Safety & Grounding |
#97
/ 113
Strong
|
#111
/ 113
Strong
|
Claude Haiku 4.5 |
| Product & Project Management |
#38
/ 110
Strong
|
#50
/ 110
Strong
|
Claude Haiku 4.5 |
| Sales |
#80
/ 110
Usable
|
#92
/ 110
Usable
|
Claude Haiku 4.5 |
| Summarization & Meeting Notes |
#27
/ 110
Excellent
|
#39
/ 110
Excellent
|
Claude Haiku 4.5 |
| Customer Support |
#99
/ 113
Usable
|
#110
/ 113
Needs editing
|
Claude Haiku 4.5 |
| Training & Education |
#36
/ 110
Excellent
|
#47
/ 110
Excellent
|
Claude Haiku 4.5 |
| Knowledge & Docs |
#71
/ 110
Usable
|
#77
/ 110
Usable
|
Claude Haiku 4.5 |
| Chef / Home Cooking |
#113
/ 126
Needs editing
|
#116
/ 126
Needs editing
|
Claude Haiku 4.5 |
| Investor & Pitch |
#63
/ 66
Needs editing
|
#65
/ 66
Needs editing
|
Claude Haiku 4.5 |
| Landing Pages |
#61
/ 72
Usable
|
#63
/ 72
Usable
|
Claude Haiku 4.5 |
Rank = position among every model config we tested in that task area (lower is better). Sorted by biggest gap first.
Same task, both models — judged
Both models answered the same test case; an independent judge graded each. Quotes are the judge's actual rationale.
Executive Assistant
Client scope creep response (Tactful Rewrite Test)
“The response generally meets the criteria but includes unrequested meta-commentary and slightly weakens the boundary by not explicitly addressing the 'for free' aspect in one option.”
“The candidate successfully rewrites the message to be professional, preserves the boundary without over-apologizing, offers clear next steps, and stays well under the word limit.”
Frequently asked
Is Claude Haiku 4.5 better than Mistral Medium 3.1?
Across 22 task areas we benchmarked, Claude Haiku 4.5 ranks higher in 21 and Mistral Medium 3.1 in 1.
Which is cheaper, Claude Haiku 4.5 or Mistral Medium 3.1?
Mistral Medium 3.1 costs 2.5× less per token ($2.4 vs $6 per 1M).
What is Claude Haiku 4.5 better at?
Claude Haiku 4.5 out-ranks Mistral Medium 3.1 at Research & Competitive Analysis, Data & Analytics, AI Strategy.
What is Mistral Medium 3.1 better at?
Mistral Medium 3.1 out-ranks Claude Haiku 4.5 at Structured Output.
More comparisons
This page is Spring Prompt, running in public
We just did this for every model. Do it for your prompt.
The rankings above come from running real tasks through real models and scoring every output. Spring Prompt is that same engine — pointed at your prompt, your test cases, and your definition of good.
- Generate test cases from your prompt — no eval set required to start.
- Compare models side by side with quality, cost and latency in one matrix.
- Optimise the winner until the scores say it's ready to ship.
Prompt × model results
12 test cases · 3 evals