Claude Opus 4.6
VS
Claude Sonnet 4.5
Claude Opus 4.6 vs Claude Sonnet 4.5: 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 Opus 4.6 wins 15 of 22 task areas we tested; Claude Sonnet 4.5 takes 7. Claude Sonnet 4.5 costs 1.7× less per token ($18 vs $30 per 1M).
Claude Sonnet 4.5 costs 1.7× less per token ($18 vs $30 per 1M).
Task by task
| Task area | Claude Opus 4.6 | Claude Sonnet 4.5 | Winner |
|---|---|---|---|
| Chef / Home Cooking |
#12
/ 126
Strong
|
#80
/ 126
Usable
|
Claude Opus 4.6 |
| AI Strategy |
#5
/ 126
Strong
|
#67
/ 126
Strong
|
Claude Opus 4.6 |
| Legal & HR |
#8
/ 110
Excellent
|
#69
/ 110
Strong
|
Claude Opus 4.6 |
| Content & Brand |
#14
/ 124
Strong
|
#74
/ 124
Usable
|
Claude Opus 4.6 |
| Creative & Comedy | #10 / 110 | #66 / 110 | Claude Opus 4.6 |
| Training & Education |
#1
/ 110
Excellent
|
#51
/ 110
Excellent
|
Claude Opus 4.6 |
| Coding |
#18
/ 115
Excellent
|
#62
/ 115
Strong
|
Claude Opus 4.6 |
| Customer Support |
#81
/ 113
Usable
|
#43
/ 113
Strong
|
Claude Sonnet 4.5 |
| Knowledge & Docs |
#66
/ 110
Usable
|
#28
/ 110
Strong
|
Claude Sonnet 4.5 |
| Investor & Pitch |
#8
/ 66
Strong
|
#42
/ 66
Usable
|
Claude Opus 4.6 |
| Summarization & Meeting Notes |
#34
/ 110
Excellent
|
#2
/ 110
Excellent
|
Claude Sonnet 4.5 |
| Product & Project Management |
#45
/ 110
Strong
|
#71
/ 110
Strong
|
Claude Opus 4.6 |
| Data & Analytics |
#10
/ 110
Excellent
|
#33
/ 110
Excellent
|
Claude Opus 4.6 |
| Translation & Localization |
#2
/ 110
Excellent
|
#25
/ 110
Excellent
|
Claude Opus 4.6 |
| Frontend & Landing Pages |
#51
/ 109
Needs editing
|
#29
/ 109
Needs editing
|
Claude Sonnet 4.5 |
| RAG, Safety & Grounding |
#59
/ 113
Excellent
|
#39
/ 113
Excellent
|
Claude Sonnet 4.5 |
| Structured Output |
#98
/ 113
Strong
|
#80
/ 113
Strong
|
Claude Sonnet 4.5 |
| Executive Assistant |
#18
/ 112
Strong
|
#31
/ 112
Strong
|
Claude Opus 4.6 |
| Presentations & Decks |
#54
/ 110
Excellent
|
#43
/ 110
Excellent
|
Claude Sonnet 4.5 |
| Research & Competitive Analysis |
#29
/ 110
Strong
|
#39
/ 110
Strong
|
Claude Opus 4.6 |
| Landing Pages |
#23
/ 72
Strong
|
#30
/ 72
Strong
|
Claude Opus 4.6 |
| Sales |
#4
/ 110
Strong
|
#10
/ 110
Strong
|
Claude Opus 4.6 |
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.
Training & Education
Student is truly stuck — give a real hint, not a non-answer (Socratic Tutoring)
“The model perfectly executes the Socratic tutoring task. It accurately identifies the foundational knowledge needed (fixed costs, variable costs, price) and provides a highly actionable, concrete first step without giving away the formula or the answer. The 'chips away' analogy for contribution margin is an excellent pedagogical tool that builds intuition before introducing math.”
“The model fails to scaffold the learning process by dumping the full break-even formula instead of guiding the student, and lacks illuminating examples.”
Frequently asked
Is Claude Opus 4.6 better than Claude Sonnet 4.5?
Across 22 task areas we benchmarked, Claude Opus 4.6 ranks higher in 15 and Claude Sonnet 4.5 in 7.
Which is cheaper, Claude Opus 4.6 or Claude Sonnet 4.5?
Claude Sonnet 4.5 costs 1.7× less per token ($18 vs $30 per 1M).
What is Claude Opus 4.6 better at?
Claude Opus 4.6 out-ranks Claude Sonnet 4.5 at Chef / Home Cooking, AI Strategy, Legal & HR.
What is Claude Sonnet 4.5 better at?
Claude Sonnet 4.5 out-ranks Claude Opus 4.6 at Customer Support, Knowledge & Docs, Summarization & Meeting Notes.
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