Claude Haiku 4.5
VS
Gemini 3.1 Pro Preview
Claude Haiku 4.5 vs Gemini 3.1 Pro Preview: 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.
Gemini 3.1 Pro Preview wins 21 of 22 task areas we tested; Claude Haiku 4.5 takes 1. Claude Haiku 4.5 costs 2.3× less per token ($6 vs $14 per 1M).
Claude Haiku 4.5 costs 2.3× less per token ($6 vs $14 per 1M).
Task by task
| Task area | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Winner |
|---|---|---|---|
| Chef / Home Cooking |
#113
/ 126
Needs editing
|
#6
/ 126
Strong
|
Gemini 3.1 Pro Preview |
| Customer Support |
#99
/ 113
Usable
|
#2
/ 113
Strong
|
Gemini 3.1 Pro Preview |
| Structured Output |
#109
/ 113
Usable
|
#15
/ 113
Excellent
|
Gemini 3.1 Pro Preview |
| RAG, Safety & Grounding |
#97
/ 113
Strong
|
#8
/ 113
Excellent
|
Gemini 3.1 Pro Preview |
| Creative & Comedy | #84 / 110 | #4 / 110 | Gemini 3.1 Pro Preview |
| Content & Brand |
#89
/ 124
Usable
|
#11
/ 124
Strong
|
Gemini 3.1 Pro Preview |
| Presentations & Decks |
#88
/ 110
Strong
|
#24
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Executive Assistant |
#73
/ 112
Strong
|
#14
/ 112
Strong
|
Gemini 3.1 Pro Preview |
| AI Strategy |
#91
/ 126
Strong
|
#33
/ 126
Strong
|
Gemini 3.1 Pro Preview |
| Knowledge & Docs |
#71
/ 110
Usable
|
#19
/ 110
Strong
|
Gemini 3.1 Pro Preview |
| Investor & Pitch |
#63
/ 66
Needs editing
|
#13
/ 66
Strong
|
Gemini 3.1 Pro Preview |
| Sales |
#80
/ 110
Usable
|
#32
/ 110
Strong
|
Gemini 3.1 Pro Preview |
| Landing Pages |
#61
/ 72
Usable
|
#14
/ 72
Strong
|
Gemini 3.1 Pro Preview |
| Training & Education |
#36
/ 110
Excellent
|
#4
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Legal & HR |
#48
/ 110
Excellent
|
#17
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Coding |
#94
/ 115
Usable
|
#64
/ 115
Strong
|
Gemini 3.1 Pro Preview |
| Frontend & Landing Pages |
#7
/ 109
Usable
|
#34
/ 109
Needs editing
|
Claude Haiku 4.5 |
| Summarization & Meeting Notes |
#27
/ 110
Excellent
|
#4
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Research & Competitive Analysis |
#59
/ 110
Usable
|
#42
/ 110
Strong
|
Gemini 3.1 Pro Preview |
| Data & Analytics |
#53
/ 110
Excellent
|
#39
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Product & Project Management |
#38
/ 110
Strong
|
#25
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Translation & Localization |
#27
/ 110
Excellent
|
#14
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
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.
Customer Support
Shipping delay (Basic Support Reply Test)
“The model failed the negative constraint by inventing refund and replacement policies not provided in the prompt. It also included meta-commentary at the end, violating the requirement to produce a ready-to-use answer.”
“The response perfectly follows the negative constraint by avoiding any invented delivery dates. It includes strong empathy, clear tracking instructions, and a logical escalation path, making it an excellent, production-ready template.”
Frequently asked
Is Claude Haiku 4.5 better than Gemini 3.1 Pro Preview?
Across 22 task areas we benchmarked, Gemini 3.1 Pro Preview ranks higher in 21 and Claude Haiku 4.5 in 1.
Which is cheaper, Claude Haiku 4.5 or Gemini 3.1 Pro Preview?
Claude Haiku 4.5 costs 2.3× less per token ($6 vs $14 per 1M).
What is Claude Haiku 4.5 better at?
Claude Haiku 4.5 out-ranks Gemini 3.1 Pro Preview at Frontend & Landing Pages.
What is Gemini 3.1 Pro Preview better at?
Gemini 3.1 Pro Preview out-ranks Claude Haiku 4.5 at Chef / Home Cooking, Customer Support, 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