Compare · ModelsLive · 2 picked · head to head
Gemma 4 26B A4B vs Llama 3.3 70B Instruct
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Gemma 4 26B A4B wins on 6/6 benchmarks
Gemma 4 26B A4B wins 6 of 6 shared benchmarks. Leads in arena · general · knowledge.
Category leads
arena·Gemma 4 26B A4B general·Gemma 4 26B A4B knowledge·Gemma 4 26B A4B math·Gemma 4 26B A4B coding·Gemma 4 26B A4B
Hype vs Reality
Attention vs performance
Gemma 4 26B A4B
#151 by perf·#7 by attention
Llama 3.3 70B Instruct
#218 by perf·#18 by attention
Best value
Gemma 4 26B A4B
1.8x better value than Llama 3.3 70B Instruct
Gemma 4 26B A4B
314.5 pts/$
$0.15/M
Llama 3.3 70B Instruct
171.9 pts/$
$0.21/M
Vendor risk
Who is behind the model
Google DeepMind
$4.20T·Tier 1
Meta AI
$1.87T·Tier 1
Head to head
6 benchmarks · 2 models
Gemma 4 26B A4B Llama 3.3 70B Instruct
Chatbot Arena Elo · Overall
Gemma 4 26B A4B leads by +119.7
Gemma 4 26B A4B
1437.5
Llama 3.3 70B Instruct
1317.8
Dtbench
Gemma 4 26B A4B leads by +25.8
Gemma 4 26B A4B
58.2
Llama 3.3 70B Instruct
32.5
GPQA diamond
Gemma 4 26B A4B leads by +34.4
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Gemma 4 26B A4B
64.3
Llama 3.3 70B Instruct
29.9
Lmca
Gemma 4 26B A4B leads by +14.3
Gemma 4 26B A4B
34.9
Llama 3.3 70B Instruct
20.6
OTIS Mock AIME 2024-2025
Gemma 4 26B A4B leads by +77.2
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Gemma 4 26B A4B
82.2
Llama 3.3 70B Instruct
5.0
WeirdML
Gemma 4 26B A4B leads by +20.7
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Gemma 4 26B A4B
35.2
Llama 3.3 70B Instruct
14.4
Full benchmark table
| Benchmark | Gemma 4 26B A4B | Llama 3.3 70B Instruct |
|---|---|---|
Chatbot Arena Elo · Overall | 1437.5 | 1317.8 |
Dtbench | 58.2 | 32.5 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 64.3 | 29.9 |
Lmca | 34.9 | 20.6 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 82.2 | 5.0 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 35.2 | 14.4 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $0.07 | $0.23 | 262K tokens (~131 books) | $1.07 | |
| $0.10 | $0.32 | 131K tokens (~66 books) | $1.55 |