Compare · ModelsLive · 3 picked · head to head
GPT-3.5 Turbo (older v0613) vs phi-3-small 7.4B vs Llama 3 8B Instruct
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-3.5 Turbo (older v0613) wins on 5/9 benchmarks
GPT-3.5 Turbo (older v0613) wins 5 of 9 shared benchmarks. Leads in knowledge · math.
Category leads
knowledge·GPT-3.5 Turbo (older v0613)reasoning·phi-3-small 7.4Bmath·GPT-3.5 Turbo (older v0613)
Hype vs Reality
Attention vs performance
GPT-3.5 Turbo (older v0613)
#111 by perf·no signal
phi-3-small 7.4B
#25 by perf·no signal
Llama 3 8B Instruct
#184 by perf·no signal
Best value
Llama 3 8B Instruct
28.8x better value than GPT-3.5 Turbo (older v0613)
GPT-3.5 Turbo (older v0613)
30.5 pts/$
$1.50/M
phi-3-small 7.4B
—
no price
Llama 3 8B Instruct
880.0 pts/$
$0.04/M
Vendor risk
Who is behind the model
OpenAI
$840.0B·Tier 1
Microsoft
$3.00T·Big Tech
Meta AI
$1.50T·Tier 1
Head to head
9 benchmarks · 3 models
GPT-3.5 Turbo (older v0613)phi-3-small 7.4BLlama 3 8B Instruct
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
phi-3-small 7.4B
37.1
Llama 3 8B Instruct
36.0
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
phi-3-small 7.4B
87.6
Llama 3 8B Instruct
77.1
MMLU
phi-3-small 7.4B leads by +9.2
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
phi-3-small 7.4B
67.6
Llama 3 8B Instruct
58.4
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
phi-3-small 7.4B
84.0
Llama 3 8B Instruct
76.8
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
phi-3-small 7.4B
58.1
Llama 3 8B Instruct
67.7
Winogrande
GPT-3.5 Turbo (older v0613) leads by +0.2
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
phi-3-small 7.4B
63.0
Llama 3 8B Instruct
51.4
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
GPQA diamond
GPT-3.5 Turbo (older v0613) leads by +1.5
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)
2.9
Llama 3 8B Instruct
1.4
MATH level 5
GPT-3.5 Turbo (older v0613) leads by +5.5
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 8B Instruct
6.1
Full benchmark table
| Benchmark | GPT-3.5 Turbo (older v0613) | phi-3-small 7.4B | Llama 3 8B Instruct |
|---|---|---|---|
ANLI ANLI (Adversarial NLI) · adversarially constructed natural language inference dataset where each round targets weaknesses found in previous model generations. | 37.1 | 37.1 | 36.0 |
ARC AI2 AI2 Reasoning Challenge · tests grade-school level science knowledge with multiple-choice questions requiring reasoning beyond simple retrieval. | 83.2 | 87.6 | 77.1 |
MMLU Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge. | 56.4 | 67.6 | 58.4 |
OpenBookQA OpenBookQA · science questions that require combining a given core fact with broad common knowledge, mimicking an open-book exam setting. | 81.3 | 84.0 | 76.8 |
TriviaQA TriviaQA · reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents. | 85.8 | 58.1 | 67.7 |
Winogrande WinoGrande · large-scale commonsense reasoning benchmark where models must resolve ambiguous pronouns in carefully constructed sentence pairs. | 63.2 | 63.0 | 51.4 |
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 | 72.1 | — |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 2.9 | — | 1.4 |
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 | — | 6.1 |
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 | |
| — | — | — | — | |
| $0.03 | $0.04 | 8K tokens (~4 books) | $0.33 |
People also compared