Which model?
Pick what you're building, tick what you can't do without, and see what fits.
What are you building?
Must have
36 models fit
Left out: 37 previous generation
For high-volume, latency-sensitive tasks such as classification, extraction, and routing
- Has a published card
Its card: “High-volume, latency-sensitive work”
$0.10 / $0.501M contextOur most efficient model for focused, high-volume tasks.
- Has a published card
Its card: “Focused, high-volume tasks at the lowest cost in the family”
$0.10 / $0.501.05M contextMinistral 3 3B is the smallest and most efficient model in the Ministral 3 family, offering robust language and vision capabilities in a compact package.
- Open weights
- Has a published card
Its card: “Image captioning”
$0.10 / $0.10256K contextThe latest multimodal model from the Qwen family, combining powerful reasoning and generation with remarkable speed
- Open weights
- Has a published card
Its card: “Agentic coding at low cost (SWE-bench Pro 62.5%, SWE-bench Multilingual 81.0%)”
$0.15 / $0.471M contextQwen's next-generation native omni-modal model
- Has a published card
Its card: “Audio and video understanding, meeting summaries and subtitles”
$0.15 / $0.471M contextMinistral 3 8B is a powerful and efficient model in the Ministral 3 family, offering best-in-class text and vision capabilities.
- Open weights
- Has a published card
Its card: “Chat interfaces in constrained environments”
$0.15 / $0.15256K contextOur powerful hybrid model unifying instruct, reasoning, and coding capabilities in a single model
- Open weights
- Has a published card
Its card: “Coding automation, codebase exploration and coding agents”
$0.15 / $0.60256K contextMinistral 3 14B is the largest model in the Ministral 3 family, offering state-of-the-art capabilities and performance comparable to its larger Mistral Small 3.2 24B counterpart.
- Open weights
- Has a published card
Its card: “Private or custom chat and assistant deployments in constrained environments”
$0.20 / $0.20256K contextSeed2.0 Lite provides a balanced trade-off for general-purpose applications
- Has a published card
Its card: “Search and deep-research agents at low cost (BrowseComp 72.1, DeepSearchQA 67.7)”
$0.25 / $2.00256K contextA low-latency, highly cost-effective subagent option designed for high-volume automation
- Has a published card
Its card: “Agentic workflows”
$0.30 / $2.501M contextOur cutting-edge language model for code completion released end of July 2025, Codestral specializes in low-latency, high-frequency tasks such as fill-in-the-middle (FIM) and code generation.
- Has a published card
Its card: “Inline code completion and multi-line edits (FIM) in IDEs, including Mistral Code for JetBrains and VS Code”
$0.30 / $0.90128K contextThe smallest model in our new architecture family, with native visual understanding
- Open weights
- Has a published card
Its card: “Agentic coding”
$0.30 / $1.201M contextThe cost-effective Plus model of the Qwen3.7 series, with multimodal interactive hybrid agent capabilities
- Has a published card
Its card: “Balanced cost and capability for chatbots, content generation, summarization and document processing”
$0.40 / $1.601M contextA 27B native vision-language dense model of the Qwen3.8 series
- Open weights
- Has a published card
Its card: “Computer use (OSWorld-Verified 84.3%) and mobile use (AndroidWorld 81.9%)”
$0.50 / $3.001M contextSeed2.1 Turbo is designed for efficient, high-throughput product scenarios
- Has a published card
Its card: “High-throughput agent products”
$0.50 / $2.50256K contextFlagship general-purpose agentic model
- Has a published card
Its card: “Search and deep-research agents (BrowseComp 77.3, HLE-text with tools 54.2)”
$0.50 / $3.00256K contextSpaceXAI's intelligent coding model for agentic software, engineering, and workflow tasks.
- Has a published card
Its card: “Agentic coding”
$1.00 / $2.00256K contextFast, reliable model with strong tool calling and instruction following capabilities.
- Has a published card
Its card: “Tool calling and instruction following”
$1.25 / $2.501M contextMuse Spark 1.3 is trained for agentic workflows and optimized for competitive coding performance.
- Has a published card
Its card: “Long-horizon agentic workflows”
$1.25 / $4.251M contextState-of-the-art, open-weight, general-purpose multimodal model
- Has a published card
Its card: “Agentic coding and agentic business workflows”
$1.36 / $4.181M contextOur most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows
- Has a published card
Its card: “Cost-effective scaling of general-purpose, production-ready agents”
$1.50 / $7.501M contextFrontier-class multimodal model optimized for agentic and coding use cases
- Open weights
- Has a published card
Its card: “Long-horizon coding and productivity work”
$1.50 / $7.50256K contextThe best combination of speed and intelligence
- Has a published card
Its card: “Agentic coding”
$2.00 / $10.001M contextNear-Astra performance for complex work at a lower cost.
- Has a published card
Its card: “Complex coding, computer use and professional work at lower cost than Astra”
$2.00 / $10.001.05M contextGoogle's most advanced model for complex tasks
- Has a published card
Its card: “Agentic performance”
$2.00 / $12.001M contextSpaceXAI's frontier model for coding, agentic tasks, and knowledge work.
- Has a published card
Its card: “Long-horizon agentic coding (CursorBench, DeepSWE, Terminal-Bench, SWE-Marathon)”
$2.00 / $6.00500K contextA 2.4-trillion-parameter MoE flagship with a major leap in coding and office productivity
- Open weights
- Has a published card
Its card: “Long-horizon agentic coding (Terminal Bench 2.1 86.6%, FrontierSWE 73.5%)”
$2.00 / $6.001M contextFor long-running agentic coding and knowledge work
- Has a published card
Its card: “Agentic coding”
$4.00 / $20.001M contextFor demanding reasoning and long-horizon agentic work
- Has a published card
Its card: “Coding, knowledge work and problem-solving, including novel mathematical and scientific reasoning”
$10.00 / $50.001M contextOur most capable model for the most demanding work.
- Has a published card
Its card: “Complex reasoning and coding”
$10.00 / $50.001.05M contextMedium-sized open-weight model for low latency
- Open weights
- Has a published card
Its card: “Lower-latency, local or specialized use cases”
Self-host128K contextMost powerful open-weight model, fits into an H100 GPU
- Open weights
- Has a published card
Its card: “Agentic workflows with strong instruction following”
Self-host128K contextOur most intelligent open models to date, purpose-built for advanced reasoning and agentic workflows
- Open weights
- Has a published card
Its card: “Text generation”
Self-host256K contextOptimized for multimodal understanding, multilingual tasks, coding, tool-calling, and powering agentic systems
- Open weights
- Has a published card
Its card: “Commercial and research use in multiple languages”
Self-host10M contextAn open model built for always-on local agents.
- Open weights
- Has a published card
Its card: “Local AI agents”
Self-host128K contextOptimized for multimodal understanding, multilingual tasks, coding, tool-calling, and powering agentic systems
- Open weights
- Has a published card
Its card: “Commercial and research use in multiple languages”
Self-host1M context
This filters on documented facts (inputs, context, price, weights, cards) and sorts them. It doesn't judge quality: test your top two or three on your own prompts.