mistral
Ministral 3 3B 2512
Ministral 3 3B 2512 is the smallest model in the Ministral 3 family from mistral, priced at $0.10 per million tokens for both input and output — the lowest output price in our tested set. It supports a 131,072-token context window and multimodal inputs (text and images). In our testing across 12 benchmarks, it ranked 41st out of 52 models overall. Despite being the entry point of the family, it earned 5/5 scores on both faithfulness and constrained rewriting — genuinely competitive results that hold up even against significantly more expensive models. For teams with extreme cost sensitivity that still need reliable source adherence and content compression, Ministral 3 3B 2512 offers a compelling floor.
Performance
In our 12-benchmark suite, Ministral 3 3B 2512's strongest scores are faithfulness, constrained rewriting, and classification. On faithfulness, it scored 5/5, tied for 1st with 32 other models out of 55 tested — meaning it reliably sticks to source material without hallucinating. On constrained rewriting, it scored 5/5, tied for 1st with just 4 other models out of 53 tested — a top-tier result. On classification, it scored 4/5, tied for 1st with 29 other models out of 53 tested. Additional 4/5 scores include tool calling (rank 18 of 54), structured output (rank 26 of 54), multilingual (rank 36 of 55), long context (rank 38 of 55), and persona consistency (rank 38 of 53). The main weaknesses are strategic analysis (2/5, rank 44 of 54) — near the bottom of the tested set — and safety calibration (1/5, rank 32 of 55). Agentic planning also scored 3/5 at rank 42 of 54.
Pricing
At $0.10 per million tokens for both input and output, Ministral 3 3B 2512 is the cheapest model in our tested set by output price. At 10 million output tokens per month, that's $1.00. At 100 million tokens, $10.00. The next step up within the family, Ministral 3 8B 2512, costs $0.15/M output ($1.50 per 10M) and scores slightly higher overall (avg 3.67 vs 3.58). Ministral 3 14B 2512 costs $0.20/M output ($2.00 per 10M) and scores notably higher (avg 3.75). For most general-purpose use cases, the 8B and 14B variants offer meaningfully better performance at modest price increments. But for high-volume, cost-constrained pipelines where faithfulness or constrained rewriting are the bottleneck, the 3B's price point is hard to beat.
mistral
Ministral 3 3B 2512
Benchmark Scores
External Benchmarks
Roles
Pricing
Input
$0.100/MTok
Output
$0.100/MTok
modelpicker.net
Real-World Costs
Pricing vs Performance
Output cost per million tokens (log scale) vs average score across our 12 internal benchmarks
Try It
from openai import OpenAI
client = OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key="YOUR_OPENROUTER_KEY",
)
response = client.chat.completions.create(
model="mistralai/ministral-3b-2512",
messages=[
{"role": "user", "content": "Hello, Ministral 3 3B 2512!"}
],
)
print(response.choices[0].message.content)Recommendation
Ministral 3 3B 2512 is the right choice for teams running high-volume pipelines where cost is the primary constraint and the tasks align with its strengths: faithfulness (sticking to source material), constrained rewriting (compression within strict length limits), and classification. RAG pipelines, editorial compression workflows, and content routing systems are natural fits — the 5/5 faithfulness score means it reliably doesn't add information that wasn't in the source. For multilingual content and structured output tasks, it performs adequately at 4/5. Avoid this model for strategic analysis (2/5, near bottom of tested set), complex reasoning, agentic workflows (3/5, rank 42 of 54), or any task requiring nuanced tradeoff analysis. Teams willing to spend an extra $0.05/M output should compare Ministral 3 8B 2512, which scores higher overall and adds a 5/5 on persona consistency.
How We Test
We test every model against our 12-benchmark suite covering tool calling, agentic planning, creative problem solving, safety calibration, and more. Each test is scored 1–5 by an LLM judge. Read our full methodology.