Ministral 3 8B 2512

Ministral 3 8B 2512 is a compact language model in Mistral AI's Ministral 3 family, positioned as a balanced, efficient option in the small and on-device category. It carries vision capabilities alongside text understanding, accepting both image and text inputs while producing text output, which makes it useful for lightweight multimodal tasks that do not require the scale of a frontier system. The model fits use cases such as high-volume API serving, edge deployment, classification, and chatbot applications where low computational overhead matters more than maximum reasoning depth.

The model operates with a 262,144-token context window, giving it ample room for extended documents and multi-turn conversations relative to its size class. Its open-weights are hosted on Hugging Face under the path mistralai/Ministral-3-8B-Instruct-2512, indicating the instruct-tuned configuration is available for local deployment. Within the Ministral 3 lineup, the 8B variant sits between smaller efficient siblings and the larger 14B model, offering a middle ground that pairs vision support with the speed and cost profile expected of a tiny language model intended for production-scale inference.

Cortecsministral-8b-2512ministral

Open provider page Provider docs

Quick Info

Provider: Cortecs
Model key: ministral-8b-2512
Release date: Dec 2, 2025
Last updated: Dec 2, 2025
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.179
Output token cost: $0.179

Limits

Output tokens: 256,000 tokens
Context window: 256,000 tokens

Transparent token rates

Compare Ministral 3 8B 2512 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Ministral 3 8B 2512

Cortecs

Coverage

Aggregator listing from Upend describes Ministral 3 8B 2512 as a balanced Mistral model supporting text and image inputs with text outputs and a 262,144 token context length, released on December 2, 2025. The model ID is mistralai/ministral-8b-2512 and it is marketed for multimodal understanding and structured output generation. Supported parameters include frequency penalty, max tokens, presence penalty, response format, seed, stop, structured outputs, temperature, tool choice, tools, and top_p, with defaults of temperature 0.3 and top_p 0.95. The page is promotional aggregator content summarizing existing model capabilities rather than a development announcement.

Cortecs

CoverageBenchmark

AI Release Tracker records Ministral 3 8B-25-12 as an open-weight release by Mistral on December 2, 2025, the same day as its 3B sibling. Verified API pricing shows $0.15 per million tokens for both input and output across Mistral's main, EU, and ZDR endpoint variants, with rates checked against mistral.ai on August 18, 2026. All listed endpoints expose a 256K context window and route through Mistral directly. The EU tier carries a higher $0.165 per million tokens rate compared to the standard and ZDR offerings, reflecting regional data residency pricing. The entry is a tracker summary confirming the underlying model release rather than new news.

Cortecs

CoverageBenchmark

ModelPicker's review profiles Ministral 3 8B 2512 with an overall score of 3.54 out of 5.00, ranking 147th in its index, and confirms $0.150 per million tokens for both input and output with a 262K context window. The model is characterized as Mistral's efficiency-oriented offering optimized for high-precision formatting and persona adherence. Capability scores highlight Constrained Rewriting and Persona Consistency at 5.0, with Structured Output, Classification, Tool Calling, Faithfulness, Long Context, and Multilingual at 4.0. Weaknesses include Safety Calibration at 1.0, Tabular Data at 2.0, and Strategic Analysis, Creative Problem Solving, and Agentic Planning at 3.0, framing it as a specialist rather than a general-purpose reasoner.

Cortecs

CoverageBenchmark

Third-party tracking page Magica documents Ministral 3 8B 2512 with a 262.1K token context window and matching $0.15 per million token input and output rates, plus a $0.01 cache read cost. The model was released on December 2, 2025, is attributed to Mistral, and is listed as open source with weights available on HuggingFace. Magica cites Artificial Analysis evaluations placing the model at 5.5 intelligence index, 9.7 coding index, and 0.6 agentic index, with a best design arena Elo of 1,067. Supported inputs include text and image with text output, function calling, and structured output, using the Mistral tokenizer. The page is an aggregator restating published specs rather than announcing a new development.

Cortecs

CoverageDocumentation

Mistral released Ministral 3 8B 2512 to general availability on December 2, 2025, under the v25.12 tag and Apache 2.0 license, as confirmed by Mistral's official documentation. The model is positioned as a powerful, efficient member of the Ministral 3 family offering best-in-class text and vision capabilities and is built for edge deployment across diverse hardware, including local setups. Its weight identifier is ministral-8b-2512+1. The GA spec lists a 256K context window with $0.15 per million tokens for both input and output in USD and EUR. Supported modalities are text plus image input and text output, exposed via /v1/chat/completions, /v1/conversations, and /v1/batch endpoints. Feature support covers structured outputs, function calling, document QnA, prefix caching, chat completions, and batching. A linked technical report accompanies the release.