Mistral Large 4 vs Step 5 Preview
Step 5 Preview leads Mistral Large 4 on coding, 61.1 to 56.4 on the Vibe Coding Index, and Step 5 Preview costs 1.5x less per output token ($2.70 vs $4.18 per million). Updated October 21, 2026; 55 benchmarks have a result for both.
Mistral Large 4 or Step 5 Preview: which to pick
Pick Step 5 Preview for coding: it scores 61.1 on the Vibe Coding Index against 56.4 for Mistral Large 4, though on 4 benchmarks against 5.
Pick Mistral Large 4 for work measured by Vibe Code Bench (Vals AI) or IOI (Vals AI), where it leads.
- Shared results
- 55
- Mistral Large 4 only
- 26
- Step 5 Preview only
- 6
- Like-for-like categories
- 8 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
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
Step 5 Preview
Step 5 Preview scores 61.1 against 56.4.
5 and 4 benchmarks behind the scores
Intelligence
AA Intelligence Index
Step 5 Preview
Step 5 Preview leads, 43.7 against 38.4.
Same board, both measured
Web design
Design Arena: websites
No call
Only Mistral Large 4 has a Design Arena: websites result, 1258.
Directional only
Long documents
Listed context window
No call
Both take 1.05M tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Step 5 Preview
Mistral Large 4 $0.0034, Step 5 Preview $0.0024 per turn.
List prices
At a glance
- Coding index
- Mistral Large 4 56.4 from 5 benchmarks, Step 5 Preview 61.1 from 4 benchmarks.
- Widest category
- Agentic work, 11.7 points apart on average over 8 shared benchmarks.
- Widest single gap
- LiveBench research: Mistral Large 4 28.1% against Step 5 Preview 72.8%, 44.7 points apart on a 0 to 100 scale.
- Output price
- Mistral Large 4 $4.18 against Step 5 Preview $2.70 per million output tokens, 1.5x apart.
- Context window
- Both take 1.05M tokens.
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 | Mistral Large 4 | Bars | Step 5 Preview | Ties |
|---|---|---|---|---|
| Coding | 2 | 10 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 2 | 6 | ||
| Reasoning | 1 | 6 | ||
| Math | 2 | 1 | ||
| Knowledge | 1 | 3 | ||
| Web design | 0 | 1 | ||
| General | 6 | 13 | ||
| All shared boards | 14 | 41 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index210
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| LiveBench coding | 60.0% | 85.6% | Step 5 Preview |
| LMArena WebDev | 1543 | 1567 | Step 5 Preview |
| Terminal-Bench 4 (Vals AI) | 27.8% | 31.8% | Step 5 Preview |
| Vibe Code Bench (Vals AI) | 78.4% | 70.4% | Mistral Large 4 |
| Code migration (Vals AI) | 30.6% | 39.0% | Step 5 Preview |
| IOI (Vals AI) | 45.3% | 38.1% | Mistral Large 4 |
| SWE-bench Multimodal (Vals AI) | 33.1% | 40.8% | Step 5 Preview |
| Vibe Code Bench 1-100 (Vals AI) | 14.3% | Not measured | Only Mistral Large 4 measured |
| SciCode | 54.2% | 58.9% | Step 5 Preview |
| FrontierCode | 39.4% | Not measured | Only Mistral Large 4 measured |
| ALE-Bench | 1213.2 | Not measured | Only Mistral Large 4 measured |
| LMArena Text: coding | 1478 | 1485 | Step 5 Preview |
| LMArena Text: software and IT | 1467 | 1473 | Step 5 Preview |
| LMArena WebDev: React | 1559 | 1583 | Step 5 Preview |
| LMArena WebDev: HTML | 1487 | 1556 | Step 5 Preview |
| LMArena WebDev: from an image | Not measured | 1550 | Only Step 5 Preview measured |
| Neither model has a result on the other 23 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 38.4 | 43.7 | Step 5 Preview |
| Epoch Capabilities Index | 150.0 | Not measured | Only Mistral Large 4 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index26
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| LiveBench research | 28.1% | 72.8% | Step 5 Preview |
| LiveBench browser | 8.0% | 24.0% | Step 5 Preview |
| LiveBench long agentic | 11.8% | 27.5% | Step 5 Preview |
| SREBench (Vals AI) | Not measured | 0.4% | Only Step 5 Preview measured |
| Finance Agent (Vals AI) | 54.7% | 50.7% | Mistral Large 4 |
| Tax Agent Bench (Vals AI) | 30.4% | 27.3% | Mistral Large 4 |
| Legal Research Bench (Vals AI) | 31.7% | 40.4% | Step 5 Preview |
| Harvey's Legal Agent Benchmark (Vals AI) | 17.5% | Not measured | Only Mistral Large 4 measured |
| Excel Modeling Benchmark (Vals AI) | 56.2% | 57.1% | Step 5 Preview |
| BioMysteryBench (Vals AI) | 67.0% | 67.4% | Step 5 Preview |
| Terminal-Bench-Science (Vals AI) | Not measured | 2.9% | Only Step 5 Preview measured |
| Vals Science Index | Not measured | 29.6% | Only Step 5 Preview measured |
| Neither model has a result on the other 29 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index16
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| LiveBench reasoning | 83.9% | 83.8% | Mistral Large 4 |
| MysteryMechanism (Vals AI) | 12.2% | 18.5% | Step 5 Preview |
| Humanity's Last Exam (Artificial Analysis run) | 35.0% | 46.5% | Step 5 Preview |
| GPQA Diamond (Epoch AI run) | 88.1% | Not measured | Only Mistral Large 4 measured |
| Chess puzzles | 33.0% | Not measured | Only Mistral Large 4 measured |
| Mystery game puzzles | 17.0% | Not measured | Only Mistral Large 4 measured |
| LMArena Text: hard prompts | 1444 | 1461 | Step 5 Preview |
| LMArena Text: expert | 1444 | 1485 | Step 5 Preview |
| LMArena Text: hard prompts in English | 1442 | 1454 | Step 5 Preview |
| LMArena Vision: diagrams | 1288 | 1308 | Step 5 Preview |
| LMArena Vision: homework | Not measured | 1265 | Only Step 5 Preview measured |
| Neither model has a result on the other 11 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index21
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| OTIS Mock AIME | 98.3% | Not measured | Only Mistral Large 4 measured |
| FrontierMath, tiers 1 to 3 | 58.6% | Not measured | Only Mistral Large 4 measured |
| FrontierMath, tier 4 | 26.8% | Not measured | Only Mistral Large 4 measured |
| ProofBench | 10.0% | 42.0% | Step 5 Preview |
| LMArena Text: math | 1469 | 1458 | Mistral Large 4 |
| LMArena Text: mathematical work | 1462 | 1460 | Mistral Large 4 |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index13
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| MedCode (Vals AI) | 40.7% | Not measured | Only Mistral Large 4 measured |
| SimpleQA Verified | 20.0% | Not measured | Only Mistral Large 4 measured |
| LMArena Text: science | 1448 | 1454 | Step 5 Preview |
| LMArena Text: medicine and healthcare | 1433 | 1439 | Step 5 Preview |
| LMArena Text: legal and government | 1425 | 1437 | Step 5 Preview |
| LMArena Text: business and finance | 1439 | 1435 | Mistral Large 4 |
| Neither model has a result on the other 12 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| Design Arena: websites | 1258 | Not measured | Only Mistral Large 4 measured |
| Design Arena: UI components | 1275 | Not measured | Only Mistral Large 4 measured |
| Design Arena: game dev | 1226 | Not measured | Only Mistral Large 4 measured |
| Design Arena: data viz | 1274 | Not measured | Only Mistral Large 4 measured |
| Design Arena: 3D | 1253 | Not measured | Only Mistral Large 4 measured |
| Design Arena: all code categories | 1259 | Not measured | Only Mistral Large 4 measured |
| Design Arena: SVG | 1233 | Not measured | Only Mistral Large 4 measured |
| Design Arena: ASCII art | 1202 | Not measured | Only Mistral Large 4 measured |
| LiveBench design | 65.6% | 76.9% | Step 5 Preview |
GeneralLike-for-likePublic boards, shown as published, kept out of the index613
| Benchmark | Mistral Large 4 | Step 5 Preview | Leader |
|---|---|---|---|
| LiveBench data analysis | 76.5% | 63.5% | Mistral Large 4 |
| Public Benefits Bench (Vals AI) | Not measured | 58.1% | Only Step 5 Preview measured |
| MedScribe (Vals AI) | 80.4% | Not measured | Only Mistral Large 4 measured |
| Long context reasoning (AA-LCR) | 81.3% | 88.3% | Step 5 Preview |
| DTBench | 88.2% | Not measured | Only Mistral Large 4 measured |
| LMCA | 35.8% | Not measured | Only Mistral Large 4 measured |
| Surface Evolver Bench | 28.1% | Not measured | Only Mistral Large 4 measured |
| Furniture assembly | 20.8% | Not measured | Only Mistral Large 4 measured |
| LMArena Text | 1431 | 1440 | Step 5 Preview |
| LMArena Text: instruction following | 1422 | 1438 | Step 5 Preview |
| LMArena Text: creative writing | 1363 | 1409 | Step 5 Preview |
| LMArena Vision | 1272 | 1265 | Mistral Large 4 |
| LMArena Text: English | 1433 | 1447 | Step 5 Preview |
| LMArena Text: non-English | 1422 | 1428 | Step 5 Preview |
| LMArena Text: Chinese | 1492 | 1500 | Step 5 Preview |
| LMArena Text: Japanese | 1415 | 1383 | Mistral Large 4 |
| LMArena Text: French | 1492 | 1438 | Mistral Large 4 |
| LMArena Text: Spanish | 1455 | 1441 | Mistral Large 4 |
| LMArena Text: Russian | 1417 | 1433 | Step 5 Preview |
| LMArena Text: multi-turn | 1435 | 1437 | Step 5 Preview |
| LMArena Text: longer queries | 1428 | 1449 | Step 5 Preview |
| LMArena Text: writing and language | 1386 | 1425 | Step 5 Preview |
| LMArena Text: entertainment and media | 1378 | 1412 | Step 5 Preview |
| LMArena Vision: English | 1283 | 1290 | Step 5 Preview |
| LMArena Vision: reading text | 1284 | 1282 | Mistral Large 4 |
| Neither model has a result on the other 21 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Mistral Large 4 | Step 5 Preview | 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 5 Preview costs less
Repository review
50K input + 3K output tokens
Step 5 Preview costs less
Agent loop
200K input + 10K output tokens
Step 5 Preview 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 | Mistral Large 4 | StepFun Step 5 Preview |
|---|---|---|
| Maker | Mistral AI | StepFun |
| Weights | No public weights listed | Open weights |
| Released | October 6, 2026 | October 8, 2026 |
| Context window | 1.05M | 1.05M |
| Input $/Mtok | $1.36 | $1.00 |
| Output $/Mtok | $4.18 | $2.70 |
| Coding index | 56.4 | 61.1 |
| Independent measurements behind it | 5 | 4 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Mistral Large 4 better than Step 5 Preview for coding?
Pick Step 5 Preview for coding: it scores 61.1 on the Vibe Coding Index against 56.4 for Mistral Large 4, though on 4 benchmarks against 5. Pick Mistral Large 4 for work measured by Vibe Code Bench (Vals AI) or IOI (Vals AI), where it leads.
Which is cheaper, Mistral Large 4 or Step 5 Preview?
Mistral Large 4 costs $1.36 per million input tokens and $4.18 per million output tokens; Step 5 Preview costs $1.00 and $2.70. Cost per solved task is known for neither of them.
Which has the bigger context window, Mistral Large 4 or Step 5 Preview?
Neither: both take 1.05M tokens.
How many benchmarks is the Mistral Large 4 vs Step 5 Preview comparison based on?
55 benchmarks have a result for both models as of October 21, 2026. Mistral Large 4's coding index rests on 5 independent benchmarks and Step 5 Preview's on 4. Where only one model has a result, the page says so instead of naming a winner.