vs
3:11
overall capability wins (0 ties across 14 scored). Llama 4 Scout leads overall.
- Reasoning2:6
- Coding1:1
- Arena0:1
- Pricing1:2
- Speed0:2
- Specs0:3
Verdict
Llama 4 Scout edges reasoning & knowledge; Llama 4 Scout ranks higher on LMArena (+23 Elo); Llama 4 Scout offers ~78.1x larger context (10M vs 128K).
- Llama 4 Scout edges reasoning & knowledge
- Llama 4 Scout ranks higher on LMArena (+23 Elo)
- Llama 4 Scout offers ~78.1x larger context (10M vs 128K)
- Llama 4 Scout allows ~8.0x more max output (33K vs 4K)
- Llama 4 Scout is ~1.3x cheaper on a blended token basis than Llama-3.1-Nemotron-70B Instruct
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Llama-3.1-Nemotron-70B Instruct
- $0.45
- Llama 4 Scout
- $0.33
Llama 4 Scout is about 1.4x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
NVIDIA
Meta
Family
Llama / Muse
Llama / Muse
License
NVIDIA Open Model
Llama 4 Community
Open weights
Yes
Yes
Release
Oct 15, 2024
Apr 5, 2025
Knowledge cutoff
-
-
API / provider
NVIDIA NIM
Together / Fireworks (ref.)
Modalities
text → text
text, image → text
Specs
Context window
128K
10M
Max output
4K
33K
Parameters
70B
109B (17B act.)
Pricing
Input $/1M
$0.35
$0.18
Output $/1M
$0.40
$0.59
Blended $/1M (3∶1)
$0.36
$0.28
Speed
tok/s
60
110
TTFT (s)
0.45
0.3
Reasoning
MMLU-Pro
69.0
74.3
GPQA Diamond
46.5
58.7
Humanity's Last Exam
4.2
3.8
AIME 2025
11.0
14.0
MATH-500
73.3
84.4
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
17.8
15.2
AA-LCR v1.1
8.3
27.7
CritPt
—
—
MMMU-Pro
—
52.9
IFBench
30.7
39.5
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
9.1
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
16.9
32.8
Terminal-Bench 2.1
—
3.7
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
38.7
—
SciCode
23.3
21.3
CursorBench
—
—
SWE-Rebench
—
6.6
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,299
1,322
| Metric | Llama-3.1-Nemotron-70B Instruct | Llama 4 Scout | Delta |
|---|---|---|---|
| Identity | |||
| Organization | NVIDIA | Meta | — |
| Family | Llama / Muse | Llama / Muse | — |
| License | NVIDIA Open Model | Llama 4 Community | — |
| Open weights | Yes | Yes | — |
| Release | Oct 15, 2024 | Apr 5, 2025 | — |
| Knowledge cutoff | - | - | — |
| API / provider | NVIDIA NIM | Together / Fireworks (ref.) | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 128K | 10M | B +9.9M |
| Max output | 4K | 33K | B +29K |
| Parameters | 70B | 109B (17B act.) | B +39 |
| Pricing | |||
| Input $/1M | $0.35 | $0.18 | B +$0.17 |
| Output $/1M | $0.40 | $0.59 | A +$0.19 |
| Blended $/1M (3∶1) | $0.36 | $0.28 | B +$0.08 |
| Speed | |||
| tok/s | 60 | 110 | B +50 |
| TTFT (s) | 0.45 | 0.3 | B +0.15 |
| Reasoning | |||
| MMLU-Pro | 69.0 | 74.3 | B +5.3 pts |
| GPQA Diamond | 46.5 | 58.7 | B +12.2 pts |
| Humanity's Last Exam | 4.2 | 3.8 | A +0.4 pts |
| AIME 2025 | 11.0 | 14.0 | B +3.0 pts |
| MATH-500 | 73.3 | 84.4 | B +11.1 pts |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 17.8 | 15.2 | A +2.6 pts |
| AA-LCR v1.1 | 8.3 | 27.7 | B +19.4 pts |
| CritPt | — | — | — |
| MMMU-Pro | — | 52.9 | — |
| IFBench | 30.7 | 39.5 | B +8.8 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | 9.1 | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 16.9 | 32.8 | B +15.9 pts |
| Terminal-Bench 2.1 | — | 3.7 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | 38.7 | — | — |
| SciCode | 23.3 | 21.3 | A +2.0 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 6.6 | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,299 | 1,322 | B +23 Elo |
Llama-3.1-Nemotron-70B Instruct
- Context
- 128K / 4K out
- Parameters
- 70B
- Price
- $0.35 / $0.40
- Speed
- 60 tok/s · 0.45s TTFT
- Modalities
- text → text
- License
- NVIDIA Open Model
NVIDIA's RLHF-tuned Llama 3.1 70B — preference-aligned open instruct model widely served on NIM.
Llama 4 Scout
- Context
- 10M / 33K out
- Parameters
- 109B (17B act.)
- Price
- $0.18 / $0.59
- Speed
- 110 tok/s · 0.3s TTFT
- Modalities
- text, image → text
- License
- Llama 4 Community
Llama 4 variant with an extreme 10M context window for long-document retrieval workloads.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.