Gemma 4 31B vs Llama 3.3 70B Instruct
Gemma 4 31B leads Llama 3.3 70B Instruct on the Artificial Analysis Intelligence Index, 14.7 to 7.7, and Gemma 4 31B costs 1.8x less per output token ($0.40 vs $0.72 per million). Updated October 22, 2026; 39 benchmarks have a result for both.
Gemma 4 31B or Llama 3.3 70B Instruct: which to pick
Only Gemma 4 31B has a coding score, 39.0 on the Vibe Coding Index; Llama 3.3 70B Instruct has none, so the two cannot be ranked on coding.
Compare them on price, context and the boards below.
- Shared results
- 39
- Gemma 4 31B only
- 32
- Llama 3.3 70B Instruct only
- 13
- Like-for-like categories
- 7 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
No call
Only Gemma 4 31B has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
Gemma 4 31B
Gemma 4 31B leads, 14.7 against 7.7.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Gemma 4 31B has a AA Agentic Index result, 4.2.
Directional only
Long documents
Listed context window
Gemma 4 31B
Gemma 4 31B takes 262K tokens against 131K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemma 4 31B
Gemma 4 31B $0.0003, Llama 3.3 70B Instruct $0.0008 per turn.
List prices
At a glance
- Coding index
- Gemma 4 31B 39.0 from 2 benchmarks, Llama 3.3 70B Instruct not benchmarked.
- Widest category
- Math, 44.6 points apart on average over 3 shared benchmarks.
- Widest single gap
- OTIS Mock AIME: Gemma 4 31B 73.3% against Llama 3.3 70B Instruct 5.1%, 68.2 points apart on a 0 to 100 scale.
- Output price
- Gemma 4 31B $0.40 against Llama 3.3 70B Instruct $0.72 per million output tokens, 1.8x apart.
- Context window
- Gemma 4 31B takes 262K tokens, Llama 3.3 70B Instruct 131K.
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 | Gemma 4 31B | Bars | Llama 3.3 70B Instruct | Ties |
|---|---|---|---|---|
| Coding | 5 | 0 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 6 | 0 | ||
| Math | 3 | 0 | ||
| Knowledge | 4 | 0 | ||
| General | 18 | 0 | ||
| All shared boards | 39 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index50
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| LiveBench coding | 49.0% | Not measured | Only Gemma 4 31B measured |
| LMArena WebDev | 1367 | Not measured | Only Gemma 4 31B measured |
| AA Coding Index | 43.4 | 11.9 | Gemma 4 31B |
| LiveCodeBench (Artificial Analysis run) | Not measured | 28.8% | Only Llama 3.3 70B Instruct measured |
| SciCode | 45.5% | Not measured | Only Gemma 4 31B measured |
| Terminal-Bench Hard (Artificial Analysis run) | 36.4% | 3.0% | Gemma 4 31B |
| WeirdML | 52.3% | 14.4% | Gemma 4 31B |
| ALE-Bench | 925.5 | Not measured | Only Gemma 4 31B measured |
| LMArena Text: coding | 1459 | 1268 | Gemma 4 31B |
| LMArena Text: software and IT | 1463 | 1273 | Gemma 4 31B |
| LMArena WebDev: React | 1359 | Not measured | Only Gemma 4 31B measured |
| LMArena WebDev: HTML | 1380 | Not measured | Only Gemma 4 31B measured |
| SciCode | 8.9% | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 26 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| AA Intelligence Index | 14.7 | 7.7 | Gemma 4 31B |
| Epoch Capabilities Index | 143.0 | 127.3 | Gemma 4 31B |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| AA Agentic Index | 4.2 | Not measured | Only Gemma 4 31B measured |
| LiveBench research | 44.4% | Not measured | Only Gemma 4 31B measured |
| LiveBench browser | 4.0% | Not measured | Only Gemma 4 31B measured |
| LiveBench long agentic | 13.3% | Not measured | Only Gemma 4 31B measured |
| τ²-Bench Telecom | 65.5% | 26.6% | Gemma 4 31B |
| BALROG | Not measured | 23.0% | Only Llama 3.3 70B Instruct measured |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 31.9% | Only Llama 3.3 70B Instruct measured |
| Neither model has a result on the other 34 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| LiveBench reasoning | 68.3% | Not measured | Only Gemma 4 31B measured |
| GPQA Diamond (Artificial Analysis run) | 85.7% | 49.8% | Gemma 4 31B |
| Humanity's Last Exam (Artificial Analysis run) | 23.6% | 3.6% | Gemma 4 31B |
| GPQA Diamond (Epoch AI run) | 75.8% | 47.4% | Gemma 4 31B |
| SimpleBench | Not measured | 19.9% | Only Llama 3.3 70B Instruct measured |
| LMArena Text: hard prompts | 1448 | 1257 | Gemma 4 31B |
| LMArena Text: expert | 1464 | 1225 | Gemma 4 31B |
| LMArena Text: hard prompts in English | 1456 | 1277 | Gemma 4 31B |
| LMArena Vision: diagrams | 1283 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: homework | 1310 | Not measured | Only Gemma 4 31B measured |
| Kagi LLM Benchmark | 63.5% | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 11 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| AA Math Index | Not measured | 7.7 | Only Llama 3.3 70B Instruct measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 7.7% | Only Llama 3.3 70B Instruct measured |
| OTIS Mock AIME | 73.3% | 5.1% | Gemma 4 31B |
| MATH level 5 | Not measured | 41.6% | Only Llama 3.3 70B Instruct measured |
| LMArena Text: math | 1463 | 1267 | Gemma 4 31B |
| LMArena Text: mathematical work | 1465 | 1267 | Gemma 4 31B |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| CaseLaw (Vals AI) | 52.6% | Not measured | Only Gemma 4 31B measured |
| MortgageTax (Vals AI) | 61.4% | Not measured | Only Gemma 4 31B measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 71.3% | Only Llama 3.3 70B Instruct measured |
| SimpleQA Verified | 10.4% | Not measured | Only Gemma 4 31B measured |
| MMLU | Not measured | 86.3% | Only Llama 3.3 70B Instruct measured |
| LMArena Text: science | 1457 | 1284 | Gemma 4 31B |
| LMArena Text: medicine and healthcare | 1451 | 1275 | Gemma 4 31B |
| LMArena Text: legal and government | 1438 | 1284 | Gemma 4 31B |
| LMArena Text: business and finance | 1430 | 1243 | Gemma 4 31B |
| FACTS Benchmark Suite | 45.3% | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 8 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index1 board, none shared
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| LiveBench design | 58.4% | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 8 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index180
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | Leader |
|---|---|---|---|
| LiveBench data analysis | 53.3% | Not measured | Only Gemma 4 31B measured |
| SAGE (Vals AI) | 55.0% | Not measured | Only Gemma 4 31B measured |
| IFBench | 75.6% | 47.1% | Gemma 4 31B |
| Long context reasoning (AA-LCR) | 69.7% | 15.7% | Gemma 4 31B |
| DTBench | 82.7% | 59.5% | Gemma 4 31B |
| LMCA | 39.3% | 17.5% | Gemma 4 31B |
| Fiction.LiveBench | Not measured | 33.3% | Only Llama 3.3 70B Instruct measured |
| Surface Evolver Bench | 30.6% | Not measured | Only Gemma 4 31B measured |
| LMArena Text | 1443 | 1274 | Gemma 4 31B |
| LMArena Text: instruction following | 1433 | 1242 | Gemma 4 31B |
| LMArena Text: creative writing | 1417 | 1250 | Gemma 4 31B |
| LMArena Vision | 1276 | Not measured | Only Gemma 4 31B measured |
| LMArena Text: English | 1451 | 1303 | Gemma 4 31B |
| LMArena Text: non-English | 1430 | 1236 | Gemma 4 31B |
| LMArena Text: Chinese | 1475 | 1217 | Gemma 4 31B |
| LMArena Text: German | Not measured | 1251 | Only Llama 3.3 70B Instruct measured |
| LMArena Text: Japanese | Not measured | 1150 | Only Llama 3.3 70B Instruct measured |
| LMArena Text: Korean | Not measured | 1142 | Only Llama 3.3 70B Instruct measured |
| LMArena Text: French | 1436 | 1282 | Gemma 4 31B |
| LMArena Text: Spanish | 1442 | 1269 | Gemma 4 31B |
| LMArena Text: Russian | 1460 | 1252 | Gemma 4 31B |
| LMArena Text: Polish | 1433 | 1259 | Gemma 4 31B |
| LMArena Text: multi-turn | 1452 | 1280 | Gemma 4 31B |
| LMArena Text: longer queries | 1446 | 1255 | Gemma 4 31B |
| LMArena Text: writing and language | 1432 | 1255 | Gemma 4 31B |
| LMArena Text: entertainment and media | 1405 | 1241 | Gemma 4 31B |
| LMArena Vision: English | 1271 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: Chinese | 1326 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: reading text | 1282 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: humor | 1280 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: captioning | 1218 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: recognizing things | 1226 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: creative writing | 1285 | Not measured | Only Gemma 4 31B measured |
| LMArena Document | 1425 | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 12 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Gemma 4 31B | Llama 3.3 70B Instruct | 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
Gemma 4 31B costs less
Repository review
50K input + 3K output tokens
Gemma 4 31B costs less
Agent loop
200K input + 10K output tokens
Gemma 4 31B 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 | Google Gemma 4 31B | Meta Llama 3.3 70B Instruct |
|---|---|---|
| Maker | Meta Llama | |
| Weights | Open weights | Open weights |
| Released | April 2, 2026 | December 6, 2024 |
| Context window | 262K | 131K |
| Input $/Mtok | $0.14 | $0.45 |
| Output $/Mtok | $0.40 | $0.72 |
| Coding index | 39.0 | Not benchmarked |
| Independent measurements behind it | 2 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Gemma 4 31B better than Llama 3.3 70B Instruct for coding?
Only Gemma 4 31B has a coding score, 39.0 on the Vibe Coding Index; Llama 3.3 70B Instruct has none, so the two cannot be ranked on coding. Compare them on price, context and the boards below.
Which is cheaper, Gemma 4 31B or Llama 3.3 70B Instruct?
Gemma 4 31B costs $0.14 per million input tokens and $0.40 per million output tokens; Llama 3.3 70B Instruct costs $0.45 and $0.72. Cost per solved task is known for neither of them.
Which has the bigger context window, Gemma 4 31B or Llama 3.3 70B Instruct?
Gemma 4 31B, with 262K tokens against 131K.
How many benchmarks is the Gemma 4 31B vs Llama 3.3 70B Instruct comparison based on?
39 benchmarks have a result for both models as of October 22, 2026. Gemma 4 31B's coding index rests on 2 independent benchmarks and Llama 3.3 70B Instruct's on 0. Where only one model has a result, the page says so instead of naming a winner.