Model comparison
Codestral vs GPT-3.5-turbo
Codestral is the stronger model overall, scoring 30.6 to 23.2 on the Noometry Index.
Last verified . 4 shared benchmarks.
Mistral AISummary
- They share 4 benchmarks with published results for both. Codestral scores higher in 2 categories and GPT-3.5-turbo in 0 categories; 2 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Codestral leads 19.8 to 13.8.
- Codestral is cheaper at $0.30 / $0.90 per million input/output tokens, against $0.50 / $1.50 for GPT-3.5-turbo.
- Codestral accepts more context: 256K tokens versus 16K.
Side by side
| Codestral | GPT-3.5-turbo | |
|---|---|---|
| Provider | Mistral AI | OpenAI |
| Noometry Index | 30.6 | 23.2 |
| Released | 2024-05-29 | 2023-03-01 |
| Weights | Proprietary | Proprietary |
| Context window | 256K | 16K |
| Max output | 8K | 4K |
| Input $ / M tokens | $0.30 | $0.50 |
| Output $ / M tokens | $0.90 | $1.50 |
| Results tracked | 7 | 44 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Codestral leads
Codestral: 27.3 (#321), GPT-3.5-turbo: 23.9 (#331)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| BigCodeBench Instruct | 41.8% | 39.1% |
| BigCodeBench Complete | 52.5% | 50.6% |
| HumanEval+ | 73.8% | 70.7% |
| MBPP+ | 61.9% | 69.7% |
| Aider Polyglot | 11.1% | — |
| WeirdML | — | 3.5% |
| LMArena Coding | — | 1136 |
| ALE-Bench | 137.78 | — |
Agentic & Tool Use Not comparable
Codestral: —, GPT-3.5-turbo: —
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| METR Time Horizons | — | 21.5% |
Reasoning Codestral leads
Codestral: 19.8 (#251), GPT-3.5-turbo: 13.8 (#332)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| Kagi LLM Benchmark | 32.5% | — |
| Chess Puzzles | — | 0% |
| LMArena Hard Prompts | — | 1108 |
| Mystery Game Puzzles | — | 3% |
| DTBench | — | 48.5% |
| LMCA | — | 9.7% |
| Adversarial NLI | — | 58.1% |
| BIG-Bench Hard | — | 61.6% |
| CommonsenseQA 2.0 | — | 57% |
| Epoch Capabilities Index | — | 118.55 |
| ForecastBench | — | 50.4 |
| WinoGrande | — | 81.6% |
Math Not comparable
Codestral: —, GPT-3.5-turbo: 6.3 (#327)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| FrontierMath (Tiers 1-3) | — | 0% |
| OTIS Mock AIME 2024-2025 | — | 2.2% |
| LMArena Math | — | 1142 |
| MATH Level 5 | — | 15.9% |
| GSM8K | — | 57.8% |
Knowledge Not comparable
Codestral: —, GPT-3.5-turbo: 10.0 (#303)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| GPQA Diamond | — | 28% |
| LMArena Expert | — | 1070 |
| ARC (AI2) Challenge | — | 87.4% |
| BoolQ | — | 87% |
| MMLU | — | 71.4% |
| OpenBookQA | — | 86% |
| TriviaQA | — | 85.8% |
Multilingual Not comparable
Codestral: —, GPT-3.5-turbo: 31.5 (#258)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| LMArena Non-English | — | 1108 |
| LMArena Chinese | — | 1075 |
| LMArena French | — | 1118 |
| LMArena German | — | 1090 |
| LMArena Japanese | — | 1043 |
| LMArena Korean | — | 1019 |
| LMArena Russian | — | 1123 |
| LMArena Spanish | — | 1121 |
Instruction Following Not comparable
Codestral: —, GPT-3.5-turbo: 57.9 (#262)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| LMArena Instruction Following | — | 1119 |
Long Context Not comparable
Codestral: —, GPT-3.5-turbo: 34.0 (#254)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| LMArena Longer Query | — | 1121 |
Writing & Preference Not comparable
Codestral: —, GPT-3.5-turbo: 25.3 (#305)
| Benchmark | Codestral | GPT-3.5-turbo |
|---|---|---|
| LMArena Text | — | 1125 |
| LMArena Creative Writing | — | 1092 |
| EQ-Bench Creative Writing | — | 451 |
| LMArena Multi-Turn | — | 1117 |
Frequently asked questions
Is Codestral better than GPT-3.5-turbo?
Codestral is the stronger model overall, scoring 30.6 to 23.2 on the Noometry Index.
Which is cheaper, Codestral or GPT-3.5-turbo?
Codestral is cheaper. It lists at $0.30 per million input tokens and $0.90 per million output tokens; GPT-3.5-turbo lists at $0.50 and $1.50.
Is Codestral or GPT-3.5-turbo better for coding?
Codestral scores higher on coding benchmarks: 27.3 versus 23.9 in the Noometry coding category.
Which has the bigger context window?
Codestral does, with 256K tokens against 16K.
How many benchmarks do Codestral and GPT-3.5-turbo share?
4 benchmarks have published results for both models. Codestral has 7 scored results on Noometry and GPT-3.5-turbo has 44.