Compare / Mistral Large 4 vs Mistral Medium 3.5
Mistral Large 4 vs Mistral Medium 3.5
Which is cheaper, which is more capable, and which does better at each kind of work, from 94 results on 6 sources that measured both models.
Of the 94 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.
Results as of 13 November 2026 (catalogue release 2026-11-13-a1171a342f96).
Mistral Large 4 or Mistral Medium 3.5? The short answer
Pick Mistral Large 4: it scores higher (Spring Prompt overall 25 against 16) and costs less. Pick Mistral Medium 3.5 only if speed matters more: it writes faster.
- Spring Prompt overall: Mistral Large 4 25, Mistral Medium 3.5 16 (out of 100; how the score works).
- Mistral Large 4 is cheaper to run: $0.003 against $0.008 per product listing on CatalogBench (2.4 times less).
- Mistral Large 4 scores higher on the Artificial Analysis Intelligence Index: 38.4 against 14.2.
- Mistral Medium 3.5 writes faster: 157 tokens a second against 60.
- Mistral Large 4 is newer: released 6 Oct 2026, against 28 Apr 2026 for Mistral Medium 3.5.
- On our own benchmarks, on the results both models have: Mistral Large 4 does better for user surveys and marketing planning; Mistral Medium 3.5 for product listings.
- On third-party benchmarks, on the results both models have: Mistral Large 4 does better for agents and tool use and reasoning and knowledge.
Head to head
Every fact side by side
| Mistral Large 4 | Mistral Medium 3.5 | |
|---|---|---|
| Price per million tokens | $1.36 in · $4.18 out | $1.50 in · $7.50 out |
| Blended (3 in : 1 out) | $2.06 | $3.00 |
| Cheapest host, blended | $1.05 (Mistral) | $3.00 (Mistral) |
| Cost per product listing (CatalogBench) | $0.003 | $0.008 |
| Cached input, per million | – | $0.15 cached input |
| Intelligence Index | 38.4 | 14.2 |
| Spring Prompt overall | 25 | 16 |
| Output speed, tokens a second | 60 | 157 |
| Context window | 1,048,576 tokens | 262,144 tokens |
| Released | 6 Oct 2026 (newer) | 28 Apr 2026 |
| Weights | Open | Closed |
| Developer based in | France | France |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.
Their slides, side by side
The same brief for both: Manchester store investment review for Hearthside Coffee, an invented company. First slide of each deck from DeckBench.
From our research
Every shared result
Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.
| Measure | Mistral Large 4 | Mistral Medium 3.5 |
|---|---|---|
| BulletBench measured by Spring Prompt · 3 results | ||
| Cost per game · Blitz 3+2US dollars, lower is better 2 · 5 of 19 measured | $0.0082 | $0.0261 |
| Ladder Elo · Lightning 10+1ladder Elo, higher is better 38 · 23 of 44 measured | 249 | 529 |
| Cost per game · Lightning 10+1US dollars, lower is better 40 · 43 of 44 measured | $0.0117 | $0.0297 |
| CatalogBench measured by Spring Prompt · 3 results | ||
| Cost per productUS dollars, lower is better 8 · 14 of 32 measured | $0.0033 | $0.0080 |
| Reliably publish-ready% of products, higher is better 27 · 25 of 31 measured | 10.7%high reasoning | 17.9% |
| Reliably publish-ready · sales brief% of products, higher is better 24 · 17 of 31 measured | 1.8%high reasoning | 10.7% |
| DeckBench measured by Spring Prompt · 3 results | ||
| Cost per deckUS dollars, lower is better 8 · 11 of 32 measured | $0.0148 | $0.0448 |
| Deck ratingrating, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. 30 · 31 of 32 measured | 622 | 612 |
| Accurate decks% of tasks, higher is better 27 · 27 of 32 measured | 0.0% | 0.0% |
| ROASBench measured by Spring Prompt · 3 results | ||
| Overall scorescore out of 100, higher is better 22 · 32 of 32 measured | 37.0 | 15.2 |
| Audience scorescore out of 100, higher is better 22 · 32 of 32 measured | 55.8 | 28.5 |
| Business scorescore out of 100, higher is better 22 · 32 of 32 measured | 34.3 | 3.4 |
| SurveyBench measured by Spring Prompt · 3 results | ||
| Analysis ratingrating, higher is betterThe rating weighs each win by how strong the other analysis was; the win rate counts wins against whichever analyses a model happened to meet, so the two can disagree. 16 · 21 of 23 measured | 796 | 598 |
| Cost per taskUS dollars, lower is better 13 · 12 of 23 measured | $0.0924 | $0.0730 |
| Analysis scorepoints of 100, higher is better 16 · 18 of 23 measured | 63.8 | 56.7 |
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better 36 · 139 of 249 measured | 38.4 | 14.2 |
| Humanity's Last Exam% of questions, higher is better 65 · 139 of 250 measured | 35.0% | 13.8% |
| Terminal-Bench 4.0% of tasks, higher is better 26 · 72 of 120 measured | 24.7% | 0.0% |
Show all 94 results (76 more)
| Measure | Mistral Large 4 | Mistral Medium 3.5 |
|---|---|---|
| BulletBench measured by Spring Prompt · 7 results | ||
| Games lost on time · Blitz 3+2% of games, lower is better 1 · 1 of 19 measured | 0.0% | 0.0% |
| Games lost on time · Lightning 10+1% of games, lower is better 36 · 1 of 44 measured | 58.3% | 0.0% |
| Invalid moves · Blitz 3+2% of moves, lower is better 18 · 16 of 19 measured | 0.6% | 0.4% |
| Invalid moves · Lightning 10+1% of moves, lower is better 42 · 1 of 44 measured | 1.8% | 0.0%no reasoning |
| Ladder Elo · Blitz 3+2ladder Elo, higher is better 15 · 15 of 19 measured | 555 | 555 |
| Median move time · Blitz 3+2milliseconds, lower is better 3 · 1 of 19 measured | 0.8 s | 0.5 s |
| Median move time · Lightning 10+1milliseconds, lower is better 30 · 6 of 44 measured | 0.8 s | 0.5 s |
| CatalogBench measured by Spring Prompt · 28 results | ||
| Channel compliance% of products, higher is better 1 · 1 of 32 measured | 100.0% | 100.0% |
| Channel rules broken% of products, lower is better 31 · 1 of 32 measured | 3.0% | 0.0% |
| Channel rules broken · sales brief% of products, lower is better 25 · 25 of 32 measured | 1.2% | 1.2% |
| Claims to check% of products, lower is better 24 · 28 of 30 measured | 0.7%high reasoning | 3.6% |
| Claims to check · sales brief% of products, lower is better 23 · 28 of 30 measured | 3.3%high reasoning | 25.6% |
| Conflicts caught% of conflicts, higher is better 1 · 31 of 32 measured | 100.0% | 90.9% |
| Content quality% of checks, higher is better 20 · 31 of 32 measured | 96.5% | 94.1% |
| Content quality · sales brief% of checks, higher is better 24 · 29 of 32 measured | 95.9% | 94.1% |
| Decision accuracy% of decisions, higher is better 29 · 30 of 32 measured | 93.1% | 92.6% |
| Failed outputs% of products, lower is better 1 · 1 of 32 measured | 0.0% | 0.0% |
| Failed outputs · sales brief% of products, lower is better 1 · 1 of 32 measured | 0.0% | 0.0% |
| Field accuracy% of missing fields, higher is better 8 · 28 of 32 measured | 94.4% | 91.7% |
| Invented values% of filled values, lower is better 28 · 1 of 32 measured | 3.5% | 0.0% |
| Missing UK information% of products, lower is better 1 · 1 of 32 measured | 0.0% | 0.0% |
| Missing UK information · sales brief% of products, lower is better 1 · 1 of 32 measured | 0.0% | 0.0% |
| Not findable% of products, lower is better 16 · 31 of 32 measured | 4.2% | 14.3% |
| Not findable · sales brief% of products, lower is better 14 · 28 of 32 measured | 3.0%high reasoning | 10.1% |
| Publish-ready listings% of products, higher is better 26 · 28 of 32 measured | 39.9%high reasoning | 29.8% |
| Publish-ready listings · sales brief% of products, higher is better 25 · 21 of 32 measured | 10.7%high reasoning | 17.3% |
| Time per productseconds, lower is better 7 · 2 of 32 measured | 8.20 | 4.90 |
| Unsupported claimsclaims per product, lower is better 22 · 30 of 32 measured | 0.91high reasoning | 1.74 |
| Unsupported claims% of products, lower is better 25 · 30 of 32 measured | 33.3%high reasoning | 49.4% |
| Unsupported claims · sales briefclaims per product, lower is better 27 · 25 of 32 measured | 3.69high reasoning | 3.55 |
| Unsupported claims · sales brief% of products, lower is better 26 · 23 of 32 measured | 80.4%high reasoning | 76.8% |
| Wrong attributes% of products, lower is better 23 · 30 of 32 measured | 19.6%high reasoning | 31.0% |
| Wrong attributes · sales brief% of products, lower is better 17 · 31 of 32 measured | 17.3%high reasoning | 36.3% |
| Wrong category or variant% of products, lower is better 4 · 32 of 32 measured | 1.2% | 11.3% |
| Wrong category or variant · sales brief% of products, lower is better 6 · 32 of 32 measured | 1.8% | 13.1% |
| DeckBench measured by Spring Prompt · 14 results | ||
| Caveat dropped% of tasks, lower is better 15 · 1 of 32 measured | 16.7% | 0.0% |
| Clean layout% of tasks, higher is better 16 · 8 of 32 measured | 33.3% | 66.7% |
| Design quality% of the maximum, higher is better 31 · 29 of 32 measured | 30.5% | 43.7% |
| Draft figure quoted% of tasks, lower is better 13 · 25 of 32 measured | 16.7% | 50.0% |
| Findings missing% of tasks, lower is better 1 · 1 of 32 measured | 0.0% | 0.0% |
| Head-to-head win rate% of comparisons, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. 31 · 30 of 32 measured | 25.0% | 25.2% |
| Layout defects% of tasks, lower is better 16 · 8 of 32 measured | 66.7% | 33.3% |
| Misleading metric used% of tasks, lower is better 1 · 1 of 32 measured | 0.0% | 0.0% |
| Presentable decks% of tasks, higher is better 15 · 15 of 32 measured | 0.0% | 0.0% |
| Recommendation late or wrong% of tasks, lower is better 29 · 24 of 32 measured | 83.3% | 50.0% |
| Slides needing work% of tasks, lower is better 32 · 16 of 32 measured | 100.0% | 50.0% |
| Time per deckseconds, lower is better 16 · 2 of 32 measured | 114 | 25.0 |
| Unsupported claims% of tasks, lower is better 29 · 24 of 32 measured | 50.0% | 33.3% |
| Unsupported numbers% of tasks, lower is better 31 · 1 of 32 measured | 33.3% | 0.0% |
| ROASBench measured by Spring Prompt · 6 results | ||
| Consistency scorescore out of 100, higher is better 26 · 31 of 32 measured | 16.7 | 8.7 |
| Contribution profitsimulated US dollars, higher is better 20 · 31 of 32 measured | $533,401 | −$159,882 |
| Cost of a runUS dollars, lower is better 9 · 13 of 32 measured | $0.0657 | $0.15 |
| Months over budgetmonths of 12, lower is better 1 · 32 of 32 measured | 0high reasoning | 3 |
| Planning scorescore out of 100, higher is better 6 · 2 of 32 measured | 55.0high reasoning | 55.2 |
| Return on ad spendprofit per $1 spent, higher is better 19 · 31 of 32 measured | $0.45 | −$0.17 |
| SurveyBench measured by Spring Prompt · 17 results | ||
| Head-to-head win rate% of comparisons, higher is betterThe rating weighs each win by how strong the other analysis was; the win rate counts wins against whichever analyses a model happened to meet, so the two can disagree. 18 · 21 of 23 measured | 35.5% | 26.3% |
| Leading or double-barrelled questions% of tasks, lower is better 9 · 17 of 23 measured | 33.3% | 55.6% |
| Mistaken next steps% of tasks, lower is better 20 · 21 of 23 measured | 66.7% | 77.8% |
| Numbers right% of answers, higher is better 16 · 20 of 23 measured | 98.9% | 75.5% |
| Objectives unanswerable% of tasks, lower is better 14 · 14 of 23 measured | 11.1% | 11.1% |
| Precision ignored% of tasks, lower is better 19 · 21 of 23 measured | 55.6% | 100.0% |
| Research scorepoints of 100, higher is better 16 · 18 of 23 measured | 71.2 | 64.2 |
| Sound research% of tasks, higher is better 14 · 14 of 23 measured | 0.0% | 0.0% |
| Stakeholder's view presumed% of tasks, lower is better 14 · 21 of 23 measured | 22.2% | 33.3% |
| Survey plan scorepoints of 100, higher is better 15 · 20 of 23 measured | 88.4 | 81.5 |
| Survey structure% of tasks, lower is better 3 · 3 of 23 measured | 11.1% | 11.1% |
| Time per taskseconds, lower is better 22 · 2 of 23 measured | 439 | 47.5 |
| Traps handled% of traps, higher is better 16 · 22 of 23 measured | 78.7% | 59.4% |
| Traps missed% of tasks, lower is better 21 · 21 of 23 measured | 100.0% | 100.0% |
| Unsupported findings% of tasks, lower is better 21 · 21 of 23 measured | 100.0% | 100.0% |
| Wrong numbers% of tasks, lower is better 15 · 21 of 23 measured | 22.2% | 100.0% |
| Wrong recommendation% of tasks, lower is better 16 · 16 of 23 measured | 44.4% | 44.4% |
| Artificial Analysis reported by Artificial Analysis · 4 results | ||
| Long-context reasoning (AA-LCR)% of questions, higher is better 32 · 120 of 241 measured | 81.3% | 69.3% |
| Output speedtokens per second, higher is better 61 · 22 of 78 measured | 60.2 | 157 |
| SciCode% of problems, higher is better 34 · 85 of 121 measured | 54.2% | 40.2% |
| Time to first answer tokenseconds, lower is better 68 · 52 of 78 measured | 33.9 | 13.3 |
Compare them with other models
Quick answers
Is Mistral Large 4 better than Mistral Medium 3.5?
Pick Mistral Large 4: it scores higher (Spring Prompt overall 25 against 16) and costs less. Pick Mistral Medium 3.5 only if speed matters more: it writes faster.
Which is cheaper, Mistral Large 4 or Mistral Medium 3.5?
Mistral Large 4 is cheaper to run: $0.003 against $0.008 per product listing on CatalogBench (2.4 times less).
Which is faster, Mistral Large 4 or Mistral Medium 3.5?
Mistral Medium 3.5 writes about 157 tokens a second on its usual API, against 60 for Mistral Large 4 (Artificial Analysis).
Which is better for product listings, Mistral Large 4 or Mistral Medium 3.5?
Mistral Medium 3.5 ranks 23 of 31 models on sale for turning a sparse product feed and photos into listings that can go live, against 26 for Mistral Large 4.
Which is better for decks from an analysis, Mistral Large 4 or Mistral Medium 3.5?
Mistral Large 4 ranks 29 of 31 models on sale for turning a finished analysis into a deck you could present as it is, against 30 for Mistral Medium 3.5.
Which is better for user surveys, Mistral Large 4 or Mistral Medium 3.5?
Mistral Large 4 ranks 16 of 22 models on sale for planning a user survey and reading its results without being misled, against 20 for Mistral Medium 3.5.
Which is better for marketing planning, Mistral Large 4 or Mistral Medium 3.5?
Mistral Large 4 ranks 22 of 31 models on sale for planning a year of ad spend without overspending, against 31 for Mistral Medium 3.5.
Which has the bigger context window, Mistral Large 4 or Mistral Medium 3.5?
Mistral Large 4: 1,048,576 tokens, against 262,144 for Mistral Medium 3.5.
Run Mistral Large 4 and Mistral Medium 3.5 on your own prompt.
Benchmarks aren't your task. Test both side by side on your prompt, with every answer checked against your rules and the cost of each one.

