Claude Sonnet 4.5 vs Step 3.5 Flash
Updated October 21, 2026. 29 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Only Claude Sonnet 4.5 has a coding score, 39.5 on the Vibe Coding Index; Step 3.5 Flash has none, so the two cannot be ranked on coding.
Step 3.5 Flash has the edge where output tokens cost 50.0x less.
- Shared results
- 29
- Claude Sonnet 4.5 only
- 63
- Step 3.5 Flash only
- 6
- Like-for-like categories
- 6 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Only Claude Sonnet 4.5 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
Claude Sonnet 4.5
Claude Sonnet 4.5 leads, 20.7 against 17.0.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Claude Sonnet 4.5 has a AA Agentic Index result, 15.8.
Directional only
Reasoning
GPQA Diamond
No call
Only Claude Sonnet 4.5 has a GPQA Diamond result, 81.6%.
Directional only
Math
AIME
No call
Only Claude Sonnet 4.5 has a AIME result, 88.2%.
Directional only
Knowledge
MMLU Pro
No call
Only Claude Sonnet 4.5 has a MMLU Pro result, 87.4%.
Directional only
Web design
Design Arena: websites
No call
Only Claude Sonnet 4.5 has a Design Arena: websites result, 1190.
Directional only
Long documents
Listed context window
Claude Sonnet 4.5
Claude Sonnet 4.5 takes 1M tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Step 3.5 Flash
Claude Sonnet 4.5 $0.011, Step 3.5 Flash $0.0003 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Claude Sonnet 4.5 has the number, $1.16.
Directional only
At a glance
- Coding index
- Claude Sonnet 4.5 39.5 from 3 benchmarks, Step 3.5 Flash not benchmarked.
- Widest category
- Reasoning, 9.1 points apart on average over 3 shared benchmarks.
- Widest single gap
- LMArena Text: creative writing: Claude Sonnet 4.5 1442 against Step 3.5 Flash 1357, 14.1 points apart on a 0 to 100 scale.
- Output price
- Claude Sonnet 4.5 $15.00 against Step 3.5 Flash $0.30 per million output tokens, 50.0x apart.
- Context window
- Claude Sonnet 4.5 takes 1M tokens, Step 3.5 Flash 262K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Claude Sonnet 4.5 | Bars | Step 3.5 Flash | Ties |
|---|---|---|---|---|
| Coding | 2 | 0 | ||
| Intelligence | 1 | 0 | ||
| Reasoning | 3 | 0 | ||
| Math | 2 | 0 | ||
| Knowledge | 4 | 0 | ||
| General | 17 | 0 | ||
| All shared boards | 29 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index20
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| SWE-bench Verified | 74.8% | Not measured | Only Claude Sonnet 4.5 measured |
| SWE-bench Multilingual | 67.0% | Not measured | Only Claude Sonnet 4.5 measured |
| LMArena WebDev | 1394 | Not measured | Only Claude Sonnet 4.5 measured |
| AA Coding Index | 52.1 | Not measured | Only Claude Sonnet 4.5 measured |
| LiveCodeBench (Vals AI) | 73.0% | Not measured | Only Claude Sonnet 4.5 measured |
| SWE-bench Verified (Vals AI) | 70.0% | Not measured | Only Claude Sonnet 4.5 measured |
| Vibe Code Bench (Vals AI) | 22.6% | Not measured | Only Claude Sonnet 4.5 measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 32.6% | Only Step 3.5 Flash measured |
| Terminal-Bench | 46.5% | Not measured | Only Claude Sonnet 4.5 measured |
| GSO-Bench | 14.7% | Not measured | Only Claude Sonnet 4.5 measured |
| WeirdML | 47.7% | Not measured | Only Claude Sonnet 4.5 measured |
| Cybench | 60.0% | Not measured | Only Claude Sonnet 4.5 measured |
| LMArena Text: coding | 1488 | 1436 | Claude Sonnet 4.5 |
| LMArena Text: software and IT | 1473 | 1434 | Claude Sonnet 4.5 |
| LMArena WebDev: React | 1386 | Not measured | Only Claude Sonnet 4.5 measured |
| LMArena WebDev: HTML | 1399 | Not measured | Only Claude Sonnet 4.5 measured |
| Neither model has a result on the other 22 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| AA Intelligence Index | 20.7 | 17.0 | Claude Sonnet 4.5 |
| Epoch Capabilities Index | 146.7 | Not measured | Only Claude Sonnet 4.5 measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index11 boards, none shared
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| AA Agentic Index | 15.8 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: agents building web apps | 1059 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: agents building full-stack apps | 1050 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: agents building mobile apps | 1137 | Not measured | Only Claude Sonnet 4.5 measured |
| τ²-Bench Telecom | Not measured | 94.4% | Only Step 3.5 Flash measured |
| METR time horizons | 67.4% | Not measured | Only Claude Sonnet 4.5 measured |
| Remote Labor Index | 2.1% | Not measured | Only Claude Sonnet 4.5 measured |
| GDPval | 42.5% | Not measured | Only Claude Sonnet 4.5 measured |
| DeepResearch Bench | 52.6% | Not measured | Only Claude Sonnet 4.5 measured |
| OSWorld | 62.9% | Not measured | Only Claude Sonnet 4.5 measured |
| Vending-Bench 2 | $3,839 | Not measured | Only Claude Sonnet 4.5 measured |
| Neither model has a result on the other 28 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| GPQA Diamond | 81.6% | Not measured | Only Claude Sonnet 4.5 measured |
| MMMU Pro (Vals AI) | 79.3% | Not measured | Only Claude Sonnet 4.5 measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 83.1% | Only Step 3.5 Flash measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 24.5% | Only Step 3.5 Flash measured |
| GPQA Diamond (Epoch AI run) | 82.3% | Not measured | Only Claude Sonnet 4.5 measured |
| ARC-AGI | 63.7% | Not measured | Only Claude Sonnet 4.5 measured |
| ARC-AGI-2 | 13.6% | Not measured | Only Claude Sonnet 4.5 measured |
| SimpleBench | 54.3% | Not measured | Only Claude Sonnet 4.5 measured |
| Chess puzzles | 10.0% | Not measured | Only Claude Sonnet 4.5 measured |
| Mystery game puzzles | 17.0% | Not measured | Only Claude Sonnet 4.5 measured |
| EBR-bench | 2.4% | Not measured | Only Claude Sonnet 4.5 measured |
| VPCT | 39.8% | Not measured | Only Claude Sonnet 4.5 measured |
| LMArena Text: hard prompts | 1462 | 1411 | Claude Sonnet 4.5 |
| LMArena Text: expert | 1483 | 1421 | Claude Sonnet 4.5 |
| LMArena Text: hard prompts in English | 1471 | 1421 | Claude Sonnet 4.5 |
| Neither model has a result on the other 6 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| AIME | 88.2% | Not measured | Only Claude Sonnet 4.5 measured |
| MGSM (Vals AI) | 94.3% | Not measured | Only Claude Sonnet 4.5 measured |
| OTIS Mock AIME | 77.8% | Not measured | Only Claude Sonnet 4.5 measured |
| FrontierMath, tiers 1 to 3 | 23.9% | Not measured | Only Claude Sonnet 4.5 measured |
| FrontierMath, tier 4 | 2.4% | Not measured | Only Claude Sonnet 4.5 measured |
| ProofBench | 19.0% | Not measured | Only Claude Sonnet 4.5 measured |
| MATH level 5 | 97.7% | Not measured | Only Claude Sonnet 4.5 measured |
| LMArena Text: math | 1449 | 1408 | Claude Sonnet 4.5 |
| LMArena Text: mathematical work | 1462 | 1421 | Claude Sonnet 4.5 |
| Neither model has a result on the other 4 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| MMLU Pro | 87.4% | Not measured | Only Claude Sonnet 4.5 measured |
| LegalBench (Vals AI) | 84.1% | Not measured | Only Claude Sonnet 4.5 measured |
| CaseLaw (Vals AI) | 62.2% | Not measured | Only Claude Sonnet 4.5 measured |
| CorpFin (Vals AI) | 62.0% | Not measured | Only Claude Sonnet 4.5 measured |
| TaxEval (Vals AI) | 73.3% | Not measured | Only Claude Sonnet 4.5 measured |
| MortgageTax (Vals AI) | 64.0% | Not measured | Only Claude Sonnet 4.5 measured |
| MedQA (Vals AI) | 94.7% | Not measured | Only Claude Sonnet 4.5 measured |
| MedCode (Vals AI) | 44.1% | Not measured | Only Claude Sonnet 4.5 measured |
| SimpleQA Verified | 30.7% | Not measured | Only Claude Sonnet 4.5 measured |
| LMArena Text: science | 1448 | 1415 | Claude Sonnet 4.5 |
| LMArena Text: medicine and healthcare | 1444 | 1408 | Claude Sonnet 4.5 |
| LMArena Text: legal and government | 1451 | 1408 | Claude Sonnet 4.5 |
| LMArena Text: business and finance | 1440 | 1394 | Claude Sonnet 4.5 |
| Neither model has a result on the other 4 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| Design Arena: websites | 1190 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: UI components | 1173 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: game dev | 1166 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: data viz | 1167 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: 3D | 1158 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: all code categories | 1183 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: SVG | 1115 | Not measured | Only Claude Sonnet 4.5 measured |
| Design Arena: ASCII art | 1204 | Not measured | Only Claude Sonnet 4.5 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index170
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 36.1% | Not measured | Only Claude Sonnet 4.5 measured |
| MedScribe (Vals AI) | 84.5% | Not measured | Only Claude Sonnet 4.5 measured |
| IFBench | Not measured | 66.5% | Only Step 3.5 Flash measured |
| Long context reasoning (AA-LCR) | Not measured | 63.0% | Only Step 3.5 Flash measured |
| DTBench | 83.2% | Not measured | Only Claude Sonnet 4.5 measured |
| LMCA | 38.8% | Not measured | Only Claude Sonnet 4.5 measured |
| LMArena Text | 1438 | 1403 | Claude Sonnet 4.5 |
| LMArena Text: instruction following | 1459 | 1385 | Claude Sonnet 4.5 |
| LMArena Text: creative writing | 1442 | 1357 | Claude Sonnet 4.5 |
| LMArena Text: English | 1450 | 1417 | Claude Sonnet 4.5 |
| LMArena Text: non-English | 1425 | 1385 | Claude Sonnet 4.5 |
| LMArena Text: Chinese | 1460 | 1446 | Claude Sonnet 4.5 |
| LMArena Text: German | 1427 | 1405 | Claude Sonnet 4.5 |
| LMArena Text: Japanese | 1390 | 1353 | Claude Sonnet 4.5 |
| LMArena Text: Korean | 1402 | 1351 | Claude Sonnet 4.5 |
| LMArena Text: French | 1459 | 1421 | Claude Sonnet 4.5 |
| LMArena Text: Spanish | 1458 | 1419 | Claude Sonnet 4.5 |
| LMArena Text: Russian | 1437 | 1385 | Claude Sonnet 4.5 |
| LMArena Text: Polish | 1423 | 1389 | Claude Sonnet 4.5 |
| LMArena Text: multi-turn | 1465 | 1404 | Claude Sonnet 4.5 |
| LMArena Text: longer queries | 1475 | 1402 | Claude Sonnet 4.5 |
| LMArena Text: writing and language | 1442 | 1372 | Claude Sonnet 4.5 |
| LMArena Text: entertainment and media | 1428 | 1360 | Claude Sonnet 4.5 |
| LMArena Document | 1450 | Not measured | Only Claude Sonnet 4.5 measured |
| Neither model has a result on the other 22 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Claude Sonnet 4.5 | Step 3.5 Flash | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Step 3.5 Flash costs less
Repository review
50K input + 3K output tokens
Step 3.5 Flash costs less
Agent loop
200K input + 10K output tokens
Step 3.5 Flash costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Anthropic Claude Sonnet 4.5 | StepFun Step 3.5 Flash |
|---|---|---|
| Maker | Anthropic | StepFun |
| Weights | No public weights listed | Open weights |
| Released | September 29, 2025 | January 29, 2026 |
| Context window | 1M | 262K |
| Input $/Mtok | $3.00 | $0.10 |
| Output $/Mtok | $15.00 | $0.30 |
| Coding index | 39.5 | Not benchmarked |
| Independent measurements behind it | 3 | 0 |
| Cost per solved task | $1.16 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Claude Sonnet 4.5 better than Step 3.5 Flash for coding?
Only Claude Sonnet 4.5 has a coding score, 39.5 on the Vibe Coding Index; Step 3.5 Flash has none, so the two cannot be ranked on coding. Step 3.5 Flash has the edge where output tokens cost 50.0x less.
Which is cheaper, Claude Sonnet 4.5 or Step 3.5 Flash?
Claude Sonnet 4.5 costs $3.00 per million input tokens and $15.00 per million output tokens; Step 3.5 Flash costs $0.10 and $0.30. Cost per solved task is known for Claude Sonnet 4.5 only.
Which has the bigger context window, Claude Sonnet 4.5 or Step 3.5 Flash?
Claude Sonnet 4.5, with 1M tokens against 262K.
How many benchmarks is the Claude Sonnet 4.5 vs Step 3.5 Flash comparison based on?
29 benchmarks have a result for both models as of October 21, 2026. Claude Sonnet 4.5's coding index rests on 3 independent benchmarks and Step 3.5 Flash's on 0. Where only one model has a result, the page says so instead of naming a winner.