Compare · ModelsLive · 3 picked · head to head
GPT-3.5 Turbo (older v0613) vs Llama 3-70B vs phi-3-small 7.4B
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Llama 3-70B wins on 6/12 benchmarks
Llama 3-70B wins 6 of 12 shared benchmarks. Leads in general · math.
Category leads
knowledge·GPT-3.5 Turbo (older v0613)reasoning·phi-3-small 7.4Bgeneral·Llama 3-70Bmath·Llama 3-70B
Hype vs Reality
Attention vs performance
GPT-3.5 Turbo (older v0613)
#251 by perf·no signal
Llama 3-70B
#258 by perf·#18 by attention
phi-3-small 7.4B
#31 by perf·#20 by attention
Best value
GPT-3.5 Turbo (older v0613)
GPT-3.5 Turbo (older v0613)
21.9 pts/$
$1.50/M
Llama 3-70B
n/a
no price
phi-3-small 7.4B
n/a
no price
Vendor risk
Who is behind the model
OpenAI
$840.0B·Tier 1
Meta AI
$1.62T·Tier 1
Microsoft
$3.87T·Big Tech
Head to head
12 benchmarks · 3 models
GPT-3.5 Turbo (older v0613)Llama 3-70Bphi-3-small 7.4B
ANLI
ANLI (Adversarial NLI) · adversarially constructed natural language inference dataset where each round targets weaknesses found in previous model generations.
GPT-3.5 Turbo (older v0613)
37.1
Llama 3-70B
36.0
phi-3-small 7.4B
37.1
ARC AI2
phi-3-small 7.4B leads by +4.4
AI2 Reasoning Challenge · tests grade-school level science knowledge with multiple-choice questions requiring reasoning beyond simple retrieval.
GPT-3.5 Turbo (older v0613)
83.2
Llama 3-70B
77.1
phi-3-small 7.4B
87.6
MMLU
Llama 3-70B leads by +4.8
Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge.
GPT-3.5 Turbo (older v0613)
56.4
Llama 3-70B
72.4
phi-3-small 7.4B
67.6
OpenBookQA
phi-3-small 7.4B leads by +2.7
OpenBookQA · science questions that require combining a given core fact with broad common knowledge, mimicking an open-book exam setting.
GPT-3.5 Turbo (older v0613)
81.3
Llama 3-70B
76.8
phi-3-small 7.4B
84.0
TriviaQA
GPT-3.5 Turbo (older v0613) leads by +18.1
TriviaQA · reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents.
GPT-3.5 Turbo (older v0613)
85.8
Llama 3-70B
67.7
phi-3-small 7.4B
58.1
Winogrande
Llama 3-70B leads by +3.8
WinoGrande · large-scale commonsense reasoning benchmark where models must resolve ambiguous pronouns in carefully constructed sentence pairs.
GPT-3.5 Turbo (older v0613)
63.2
Llama 3-70B
67.0
phi-3-small 7.4B
63.0
BBH
phi-3-small 7.4B leads by +23.3
BIG-Bench Hard · a curated subset of 23 challenging tasks from BIG-Bench where language models previously failed to outperform average humans.
GPT-3.5 Turbo (older v0613)
48.8
phi-3-small 7.4B
72.1
Chess Puzzles
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
GPT-3.5 Turbo (older v0613)
0.0
Llama 3-70B
0.0
Dtbench
Llama 3-70B leads by +9.4
GPT-3.5 Turbo (older v0613)
14.2
Llama 3-70B
23.6
GPQA diamond
Llama 3-70B leads by +18.6
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
GPT-3.5 Turbo (older v0613)
3.0
Llama 3-70B
21.6
MATH level 5
Llama 3-70B leads by +10.9
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
GPT-3.5 Turbo (older v0613)
11.6
Llama 3-70B
22.6
OTIS Mock AIME 2024-2025
Llama 3-70B leads by +2.1
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
GPT-3.5 Turbo (older v0613)
2.1
Llama 3-70B
4.2
Full benchmark table
| Benchmark | GPT-3.5 Turbo (older v0613) | Llama 3-70B | phi-3-small 7.4B |
|---|---|---|---|
ANLI ANLI (Adversarial NLI) · adversarially constructed natural language inference dataset where each round targets weaknesses found in previous model generations. | 37.1 | 36.0 | 37.1 |
ARC AI2 AI2 Reasoning Challenge · tests grade-school level science knowledge with multiple-choice questions requiring reasoning beyond simple retrieval. | 83.2 | 77.1 | 87.6 |
MMLU Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge. | 56.4 | 72.4 | 67.6 |
OpenBookQA OpenBookQA · science questions that require combining a given core fact with broad common knowledge, mimicking an open-book exam setting. | 81.3 | 76.8 | 84.0 |
TriviaQA TriviaQA · reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents. | 85.8 | 67.7 | 58.1 |
Winogrande WinoGrande · large-scale commonsense reasoning benchmark where models must resolve ambiguous pronouns in carefully constructed sentence pairs. | 63.2 | 67.0 | 63.0 |
BBH BIG-Bench Hard · a curated subset of 23 challenging tasks from BIG-Bench where language models previously failed to outperform average humans. | 48.8 | n/a | 72.1 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 0.0 | 0.0 | n/a |
Dtbench | 14.2 | 23.6 | n/a |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 3.0 | 21.6 | n/a |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 11.6 | 22.6 | n/a |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 2.1 | 4.2 | n/a |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $1.00 | $2.00 | 4K tokens (~2 books) | $12.50 | |
| n/a | n/a | n/a | n/a | |
| n/a | n/a | n/a | n/a |
People also compared