Alternatives to Devstral Small 2
Compare Devstral Small 2 alternatives for your business or organization using the curated list below. SourceForge ranks the best alternatives to Devstral Small 2 in 2026. Compare features, ratings, user reviews, pricing, and more from Devstral Small 2 competitors and alternatives in order to make an informed decision for your business.
-
1
DeepCoder
Agentica Project
DeepCoder is a fully open source code-reasoning and generation model released by Agentica Project in collaboration with Together AI. It is fine-tuned from DeepSeek-R1-Distilled-Qwen-14B using distributed reinforcement learning, achieving a 60.6% accuracy on LiveCodeBench (representing an 8% improvement over the base), a performance level that matches that of proprietary models such as o3-mini (2025-01-031 Low) and o1 while using only 14 billion parameters. It was trained over 2.5 weeks on 32 H100 GPUs with a curated dataset of roughly 24,000 coding problems drawn from verified sources (including TACO-Verified, PrimeIntellect SYNTHETIC-1, and LiveCodeBench submissions), each problem requiring a verifiable solution and at least five unit tests to ensure reliability for RL training. To handle long-range context, DeepCoder employs techniques such as iterative context lengthening and overlong filtering.Starting Price: Free -
2
DeepSWE
Agentica Project
DeepSWE is a fully open source, state-of-the-art coding agent built on top of the Qwen3-32B foundation model and trained exclusively via reinforcement learning (RL), without supervised finetuning or distillation from proprietary models. It is developed using rLLM, Agentica’s open source RL framework for language agents. DeepSWE operates as an agent; it interacts with a simulated development environment (via the R2E-Gym environment) using a suite of tools (file editor, search, shell-execution, submit/finish), enabling it to navigate codebases, edit multiple files, compile/run tests, and iteratively produce patches or complete engineering tasks. DeepSWE exhibits emergent behaviors beyond simple code generation; when presented with bugs or feature requests, the agent reasons about edge cases, seeks existing tests in the repository, proposes patches, writes extra tests for regressions, and dynamically adjusts its “thinking” effort.Starting Price: Free -
3
Nemotron 3 Nano
NVIDIA
Nemotron 3 Nano is the smallest model in the NVIDIA Nemotron 3 family, built for agentic AI applications with strong reasoning, conversational ability, and cost-efficient inference. It is a hybrid Mamba-Transformer Mixture-of-Experts model with 3.2 billion active parameters, 3.6 billion including embeddings, and 31.6 billion total parameters. NVIDIA describes it as more accurate than the previous Nemotron 2 Nano while activating less than half of the parameters per forward pass, improving efficiency without sacrificing performance. The model is positioned as more accurate than GPT-OSS-20B and Qwen3-30B-A3B-Thinking-2507 on popular benchmarks across different categories. On an 8K input and 16K output setting using a single H200, it delivers inference throughput 3.3 times higher than Qwen3-30B-A3B and 2.2 times higher than GPT-OSS-20B. Nemotron 3 Nano supports context lengths up to 1 million tokens and is reported to outperform GPT-OSS-20B and Qwen3-30B-A3B-Instruct-2507. -
4
Devstral 2
Mistral AI
Devstral 2 is a next-generation, open source agentic AI model tailored for software engineering: it doesn’t just suggest code snippets, it understands and acts across entire codebases, enabling multi-file edits, bug fixes, refactoring, dependency resolution, and context-aware code generation. The Devstral 2 family includes a large 123-billion-parameter model as well as a smaller 24-billion-parameter variant (“Devstral Small 2”), giving teams flexibility; the larger model excels in heavy-duty coding tasks requiring deep context, while the smaller one can run on more modest hardware. With a vast context window of up to 256 K tokens, Devstral 2 can reason across extensive repositories, track project history, and maintain a consistent understanding of lengthy files, an advantage for complex, real-world projects. The CLI tracks project metadata, Git statuses, and directory structure to give the model context, making “vibe-coding” more powerful.Starting Price: Free -
5
Devstral
Mistral AI
Devstral is an open source, agentic large language model (LLM) developed by Mistral AI in collaboration with All Hands AI, specifically designed for software engineering tasks. It excels at navigating complex codebases, editing multiple files, and resolving real-world issues, outperforming all open source models on the SWE-Bench Verified benchmark with a score of 46.8%. Devstral is fine-tuned from Mistral-Small-3.1 and features a long context window of up to 128,000 tokens. It is optimized for local deployment on high-end hardware, such as a Mac with 32GB RAM or an Nvidia RTX 4090 GPU, and is compatible with inference frameworks like vLLM, Transformers, and Ollama. Released under the Apache 2.0 license, Devstral is available for free and can be accessed via Hugging Face, Ollama, Kaggle, Unsloth, and LM Studio.Starting Price: $0.1 per million input tokens -
6
GPT-5.2-Codex
OpenAI
GPT-5.2-Codex is OpenAI’s most advanced agentic coding model, built for complex, real-world software engineering and defensive cybersecurity work. It is a specialized version of GPT-5.2 optimized for long-horizon coding tasks such as large refactors, migrations, and feature development. The model maintains full context over extended sessions through native context compaction. GPT-5.2-Codex delivers state-of-the-art performance on benchmarks like SWE-Bench Pro and Terminal-Bench 2.0. It operates reliably across large repositories and native Windows environments. Stronger vision capabilities allow it to interpret screenshots, diagrams, and UI designs during development. GPT-5.2-Codex is designed to be a dependable partner for professional engineering workflows. -
7
Stable LM
Stability AI
Stable LM: Stability AI Language Models. The release of Stable LM builds on our experience in open-sourcing earlier language models with EleutherAI, a nonprofit research hub. These language models include GPT-J, GPT-NeoX, and the Pythia suite, which were trained on The Pile open-source dataset. Many recent open-source language models continue to build on these efforts, including Cerebras-GPT and Dolly-2. Stable LM is trained on a new experimental dataset built on The Pile, but three times larger with 1.5 trillion tokens of content. We will release details on the dataset in due course. The richness of this dataset gives Stable LM surprisingly high performance in conversational and coding tasks, despite its small size of 3 to 7 billion parameters (by comparison, GPT-3 has 175 billion parameters). Stable LM 3B is a compact language model designed to operate on portable digital devices like handhelds and laptops, and we’re excited about its capabilities and portability.Starting Price: Free -
8
Qwen3-Coder
Qwen
Qwen3‑Coder is an agentic code model available in multiple sizes, led by the 480B‑parameter Mixture‑of‑Experts variant (35B active) that natively supports 256K‑token contexts (extendable to 1M) and achieves state‑of‑the‑art results comparable to Claude Sonnet 4. Pre‑training on 7.5T tokens (70 % code) and synthetic data cleaned via Qwen2.5‑Coder optimized both coding proficiency and general abilities, while post‑training employs large‑scale, execution‑driven reinforcement learning, scaling test‑case generation for diverse coding challenges, and long‑horizon RL across 20,000 parallel environments to excel on multi‑turn software‑engineering benchmarks like SWE‑Bench Verified without test‑time scaling. Alongside the model, the open source Qwen Code CLI (forked from Gemini Code) unleashes Qwen3‑Coder in agentic workflows with customized prompts, function calling protocols, and seamless integration with Node.js, OpenAI SDKs, and environment variables.Starting Price: Free -
9
Claude Opus 4.5
Anthropic
Claude Opus 4.5 is Anthropic’s newest flagship model, delivering major improvements in reasoning, coding, agentic workflows, and real-world problem solving. It outperforms previous models and leading competitors on benchmarks such as SWE-bench, multilingual coding tests, and advanced agent evaluations. Opus 4.5 also introduces stronger safety features, including significantly higher resistance to prompt injection and improved alignment across sensitive tasks. Developers gain new controls through the Claude API—like effort parameters, context compaction, and advanced tool use—allowing for more efficient, longer-running agentic workflows. Product updates across Claude, Claude Code, the Chrome extension, and Excel integrations expand how users interact with the model for software engineering, research, and everyday productivity. Overall, Claude Opus 4.5 marks a substantial step forward in capability, reliability, and usability for developers, enterprises, and end users. -
10
GLM-4.1V
Zhipu AI
GLM-4.1V is a vision-language model, providing a powerful, compact multimodal model designed for reasoning and perception across images, text, and documents. The 9-billion-parameter variant (GLM-4.1V-9B-Thinking) is built on the GLM-4-9B foundation and enhanced through a specialized training paradigm using Reinforcement Learning with Curriculum Sampling (RLCS). It supports a 64k-token context window and accepts high-resolution inputs (up to 4K images, any aspect ratio), enabling it to handle complex tasks such as optical character recognition, image captioning, chart and document parsing, video and scene understanding, GUI-agent workflows (e.g., interpreting screenshots, recognizing UI elements), and general vision-language reasoning. In benchmark evaluations at the 10 B-parameter scale, GLM-4.1V-9B-Thinking achieved top performance on 23 of 28 tasks.Starting Price: Free -
11
Solar Pro 2
Upstage AI
Solar Pro 2 is Upstage’s latest frontier‑scale large language model, designed to power complex tasks and agent‑like workflows across domains such as finance, healthcare, and legal. Packaged in a compact 31 billion‑parameter architecture, it delivers top‑tier multilingual performance, especially in Korean, where it outperforms much larger models on benchmarks like Ko‑MMLU, Hae‑Rae, and Ko‑IFEval, while also excelling in English and Japanese. Beyond superior language understanding and generation, Solar Pro 2 offers next‑level intelligence through an advanced Reasoning Mode that significantly boosts multi‑step task accuracy on challenges ranging from general reasoning (MMLU, MMLU‑Pro, HumanEval) to complex mathematics (Math500, AIME) and software engineering (SWE‑Bench Agentless), achieving problem‑solving efficiency comparable to or exceeding that of models twice its size. Enhanced tool‑use capabilities enable the model to interact seamlessly with external APIs and data sources.Starting Price: $0.1 per 1M tokens -
12
Mistral NeMo
Mistral AI
Mistral NeMo, our new best small model. A state-of-the-art 12B model with 128k context length, and released under the Apache 2.0 license. Mistral NeMo is a 12B model built in collaboration with NVIDIA. Mistral NeMo offers a large context window of up to 128k tokens. Its reasoning, world knowledge, and coding accuracy are state-of-the-art in its size category. As it relies on standard architecture, Mistral NeMo is easy to use and a drop-in replacement in any system using Mistral 7B. We have released pre-trained base and instruction-tuned checkpoints under the Apache 2.0 license to promote adoption for researchers and enterprises. Mistral NeMo was trained with quantization awareness, enabling FP8 inference without any performance loss. The model is designed for global, multilingual applications. It is trained on function calling and has a large context window. Compared to Mistral 7B, it is much better at following precise instructions, reasoning, and handling multi-turn conversations.Starting Price: Free -
13
Ministral 3
Mistral AI
Mistral 3 is the latest generation of open-weight AI models from Mistral AI, offering a full family of models, from small, edge-optimized versions to a flagship, large-scale multimodal model. The lineup includes three compact “Ministral 3” models (3B, 8B, and 14B parameters) designed for efficiency and deployment on constrained hardware (even laptops, drones, or edge devices), plus the powerful “Mistral Large 3,” a sparse mixture-of-experts model with 675 billion total parameters (41 billion active). The models support multimodal and multilingual tasks, not only text, but also image understanding, and have demonstrated best-in-class performance on general prompts, multilingual conversations, and multimodal inputs. The base and instruction-fine-tuned versions are released under the Apache 2.0 license, enabling broad customization and integration in enterprise and open source projects.Starting Price: Free -
14
Magistral
Mistral AI
Magistral is Mistral AI’s first reasoning‑focused language model family, released in two sizes: Magistral Small, a 24 B‑parameter open‑weight model under Apache 2.0 (downloadable on Hugging Face), and Magistral Medium, a more capable enterprise version available via Mistral’s API, Le Chat platform, and major cloud marketplaces. Built for domain‑specific, transparent, multilingual reasoning across tasks like math, physics, structured calculations, programmatic logic, decision trees, and rule‑based systems, Magistral produces chain‑of‑thought outputs in the user’s language that you can follow and verify. This launch marks a shift toward compact yet powerful transparent AI reasoning. Magistral Medium is currently available in preview on Le Chat, the API, SageMaker, WatsonX, Azure AI, and Google Cloud Marketplace. Magistral is ideal for general-purpose use requiring longer thought processing and better accuracy than with non-reasoning LLMs. -
15
Reka Flash 3
Reka
Reka Flash 3 is a 21-billion-parameter multimodal AI model developed by Reka AI, designed to excel in general chat, coding, instruction following, and function calling. It processes and reasons with text, images, video, and audio inputs, offering a compact, general-purpose solution for various applications. Trained from scratch on diverse datasets, including publicly accessible and synthetic data, Reka Flash 3 underwent instruction tuning on curated, high-quality data to optimize performance. The final training stage involved reinforcement learning using REINFORCE Leave One-Out (RLOO) with both model-based and rule-based rewards, enhancing its reasoning capabilities. With a context length of 32,000 tokens, Reka Flash 3 performs competitively with proprietary models like OpenAI's o1-mini, making it suitable for low-latency or on-device deployments. The model's full precision requires 39GB (fp16), but it can be compressed to as small as 11GB using 4-bit quantization. -
16
EXAONE Deep
LG
EXAONE Deep is a series of reasoning-enhanced language models developed by LG AI Research, featuring parameter sizes of 2.4 billion, 7.8 billion, and 32 billion. These models demonstrate superior capabilities in various reasoning tasks, including math and coding benchmarks. Notably, EXAONE Deep 2.4B outperforms other models of comparable size, EXAONE Deep 7.8B surpasses both open-weight models of similar scale and the proprietary reasoning model OpenAI o1-mini, and EXAONE Deep 32B shows competitive performance against leading open-weight models. The repository provides comprehensive documentation covering performance evaluations, quickstart guides for using EXAONE Deep models with Transformers, explanations of quantized EXAONE Deep weights in AWQ and GGUF formats, and instructions for running EXAONE Deep models locally using frameworks like llama.cpp and Ollama.Starting Price: Free -
17
Mistral Large 3
Mistral AI
Mistral Large 3 is a next-generation, open multimodal AI model built with a powerful sparse Mixture-of-Experts architecture featuring 41B active parameters out of 675B total. Designed from scratch on NVIDIA H200 GPUs, it delivers frontier-level reasoning, multilingual performance, and advanced image understanding while remaining fully open-weight under the Apache 2.0 license. The model achieves top-tier results on modern instruction benchmarks, positioning it among the strongest permissively licensed foundation models available today. With native support across vLLM, TensorRT-LLM, and major cloud providers, Mistral Large 3 offers exceptional accessibility and performance efficiency. Its design enables enterprise-grade customization, letting teams fine-tune or adapt the model for domain-specific workflows and proprietary applications. Mistral Large 3 represents a major advancement in open AI, offering frontier intelligence without sacrificing transparency or control.Starting Price: Free -
18
Composer 2
Cursor
Composer 2 is an advanced AI coding model integrated into Cursor, designed to deliver high-level programming performance at a cost-efficient price. It is trained on long-horizon coding tasks, enabling it to solve complex problems that require multiple steps and actions. The model demonstrates strong improvements across key benchmarks, including Terminal-Bench and SWE-bench Multilingual. With enhanced intelligence and efficiency, it provides faster and more accurate code generation. Composer 2 combines strong performance with affordable pricing, making it accessible for developers and teams.Starting Price: $0.50/M input -
19
Qwen3-Coder-Next
Alibaba
Qwen3-Coder-Next is an open-weight language model specifically designed for coding agents and local development that delivers advanced coding reasoning, complex tool usage, and robust performance on long-horizon programming tasks with high efficiency, using a mixture-of-experts architecture that balances powerful capabilities with resource-friendly operation. It provides enhanced agentic coding abilities that help software developers, AI system builders, and automated coding workflows generate, debug, and reason about code with deep contextual understanding while recovering from execution errors, making it well-suited for autonomous coding agents and development-oriented applications. By achieving strong performance comparable to much larger parameter models while requiring fewer active parameters, Qwen3-Coder-Next enables cost-effective deployment for dynamic and complex programming workloads in research and production environments.Starting Price: Free -
20
Pixtral Large
Mistral AI
Pixtral Large is a 124-billion-parameter open-weight multimodal model developed by Mistral AI, building upon their Mistral Large 2 architecture. It integrates a 123-billion-parameter multimodal decoder with a 1-billion-parameter vision encoder, enabling advanced understanding of documents, charts, and natural images while maintaining leading text comprehension capabilities. With a context window of 128,000 tokens, Pixtral Large can process at least 30 high-resolution images simultaneously. The model has demonstrated state-of-the-art performance on benchmarks such as MathVista, DocVQA, and VQAv2, surpassing models like GPT-4o and Gemini-1.5 Pro. Pixtral Large is available under the Mistral Research License for research and educational use, and under the Mistral Commercial License for commercial applications.Starting Price: Free -
21
Qwen3.6
Alibaba
Qwen3.6 is a large language model developed by Alibaba as part of its Qwen AI model family, designed for real-world applications and advanced reasoning tasks. It focuses on improving stability, usability, and performance compared to earlier versions. The model supports multimodal capabilities, allowing it to process and reason across text, images, and other data types. Qwen3.6 is particularly strong in coding and developer workflows, offering improved accuracy for complex programming tasks. It uses a mixture-of-experts architecture, enabling efficient performance while maintaining large-scale model capabilities. The model is designed to be deployable in production environments, including enterprise and cloud-based systems. It can be integrated into applications or run locally using open-weight variants. Overall, Qwen3.6 delivers a powerful, efficient, and versatile AI solution for modern use cases.Starting Price: Free -
22
GLM-4.7
Zhipu AI
GLM-4.7 is an advanced large language model designed to significantly elevate coding, reasoning, and agentic task performance. It delivers major improvements over GLM-4.6 in multilingual coding, terminal-based tasks, and real-world software engineering benchmarks such as SWE-bench and Terminal Bench. GLM-4.7 supports “thinking before acting,” enabling more stable, accurate, and controllable behavior in complex coding and agent workflows. The model also introduces strong gains in UI and frontend generation, producing cleaner webpages, better layouts, and more polished slides. Enhanced tool-using capabilities allow GLM-4.7 to perform more effectively in web browsing, automation, and agent benchmarks. Its reasoning and mathematical performance has improved substantially, showing strong results on advanced evaluation suites. GLM-4.7 is available via Z.ai, API platforms, coding agents, and local deployment for flexible adoption.Starting Price: Free -
23
Mistral Medium 3
Mistral AI
Mistral Medium 3 is a powerful AI model designed to deliver state-of-the-art performance at a fraction of the cost compared to other models. It offers simpler deployment options, allowing for hybrid or on-premises configurations. Mistral Medium 3 excels in professional applications like coding and multimodal understanding, making it ideal for enterprise use. Its low-cost structure makes it highly accessible while maintaining top-tier performance, outperforming many larger models in specific domains.Starting Price: Free -
24
GLM-4.5V-Flash
Zhipu AI
GLM-4.5V-Flash is an open source vision-language model, designed to bring strong multimodal capabilities into a lightweight, deployable package. It supports image, video, document, and GUI inputs, enabling tasks such as scene understanding, chart and document parsing, screen reading, and multi-image analysis. Compared to larger models in the series, GLM-4.5V-Flash offers a compact footprint while retaining core VLM capabilities like visual reasoning, video understanding, GUI task handling, and complex document parsing. It can serve in “GUI agent” workflows, meaning it can interpret screenshots or desktop captures, recognize icons or UI elements, and assist with automated desktop or web-based tasks. Although it forgoes some of the largest-model performance gains, GLM-4.5V-Flash remains versatile for real-world multimodal tasks where efficiency, lower resource usage, and broad modality support are prioritized.Starting Price: Free -
25
Claude Sonnet 4
Anthropic
Claude Sonnet 4, the latest evolution of Anthropic’s language models, offers a significant upgrade in coding, reasoning, and performance. Designed for diverse use cases, Sonnet 4 builds upon the success of its predecessor, Claude Sonnet 3.7, delivering more precise responses and better task execution. With a state-of-the-art 72.7% performance on the SWE-bench, it stands out in agentic scenarios, offering enhanced steerability and clear reasoning capabilities. Whether handling software development, multi-feature app creation, or complex problem-solving, Claude Sonnet 4 ensures higher code quality, reduced errors, and a smoother development process.Starting Price: $3 / 1 million tokens (input) -
26
GPT-4.1 nano
OpenAI
GPT-4.1 nano is the smallest and most efficient version of OpenAI's GPT-4.1 model, optimized for low-latency, cost-effective AI processing. Despite its compact size, GPT-4.1 nano delivers strong performance with a 1 million token context window, making it ideal for applications like classification, autocompletion, and smaller-scale tasks that require fast responses. It provides a highly efficient solution for businesses and developers who need an AI model that balances speed, cost, and performance.Starting Price: $0.10 per 1M tokens (input) -
27
Claude Opus 4.1
Anthropic
Claude Opus 4.1 is an incremental upgrade to Claude Opus 4 that boosts coding, agentic reasoning, and data-analysis performance without changing deployment complexity. It raises coding accuracy to 74.5 percent on SWE-bench Verified and sharpens in-depth research and detailed tracking for agentic search tasks. GitHub reports notable gains in multi-file code refactoring, while Rakuten Group highlights its precision in pinpointing exact corrections within large codebases without introducing bugs. Independent benchmarks show about a one-standard-deviation improvement on junior developer tests compared to Opus 4, mirroring major leaps seen in prior Claude releases. -
28
OpenAI o1-mini
OpenAI
OpenAI o1-mini is a new, cost-effective AI model designed for enhanced reasoning, particularly excelling in STEM fields like mathematics and coding. It's part of the o1 series, which focuses on solving complex problems by spending more time "thinking" through solutions. Despite being smaller and 80% cheaper than its sibling, the o1-preview, o1-mini performs competitively in coding tasks and mathematical reasoning, making it an accessible option for developers and enterprises looking for efficient AI solutions. -
29
Qwen2
Alibaba
Qwen2 is the large language model series developed by Qwen team, Alibaba Cloud. Qwen2 is a series of large language models developed by the Qwen team at Alibaba Cloud. It includes both base language models and instruction-tuned models, ranging from 0.5 billion to 72 billion parameters, and features both dense models and a Mixture-of-Experts model. The Qwen2 series is designed to surpass most previous open-weight models, including its predecessor Qwen1.5, and to compete with proprietary models across a broad spectrum of benchmarks in language understanding, generation, multilingual capabilities, coding, mathematics, and reasoning.Starting Price: Free -
30
Qwen3.6-27B
Alibaba
Qwen3.6-27B is a dense, open source multimodal language model in the Qwen3.6 series, designed to deliver flagship-level performance in coding, reasoning, and agent-based workflows while maintaining a relatively efficient parameter size of 27 billion. It is positioned as a high-performance general model that “punches above its weight,” achieving results competitive with or superior to significantly larger models on key benchmarks, particularly in agentic coding tasks. It supports both thinking and non-thinking modes, allowing it to dynamically balance deep reasoning with fast responses depending on the task, and integrates capabilities across text and multimodal inputs such as images and video. Built as part of the Qwen3.6 family, the model emphasizes real-world usability, stability, and developer productivity, incorporating improvements driven by community feedback and practical deployment needs.Starting Price: Free -
31
LongLLaMA
LongLLaMA
This repository contains the research preview of LongLLaMA, a large language model capable of handling long contexts of 256k tokens or even more. LongLLaMA is built upon the foundation of OpenLLaMA and fine-tuned using the Focused Transformer (FoT) method. LongLLaMA code is built upon the foundation of Code Llama. We release a smaller 3B base variant (not instruction tuned) of the LongLLaMA model on a permissive license (Apache 2.0) and inference code supporting longer contexts on hugging face. Our model weights can serve as the drop-in replacement of LLaMA in existing implementations (for short context up to 2048 tokens). Additionally, we provide evaluation results and comparisons against the original OpenLLaMA models.Starting Price: Free -
32
Tülu 3
Ai2
Tülu 3 is an advanced instruction-following language model developed by the Allen Institute for AI (Ai2), designed to enhance capabilities in areas such as knowledge, reasoning, mathematics, coding, and safety. Built upon the Llama 3 Base, Tülu 3 employs a comprehensive four-stage post-training process: meticulous prompt curation and synthesis, supervised fine-tuning on a diverse set of prompts and completions, preference tuning using both off- and on-policy data, and a novel reinforcement learning approach to bolster specific skills with verifiable rewards. This open-source model distinguishes itself by providing full transparency, including access to training data, code, and evaluation tools, thereby closing the performance gap between open and proprietary fine-tuning methods. Evaluations indicate that Tülu 3 outperforms other open-weight models of similar size, such as Llama 3.1-Instruct and Qwen2.5-Instruct, across various benchmarks.Starting Price: Free -
33
DeepScaleR
Agentica Project
DeepScaleR is a 1.5-billion-parameter language model fine-tuned from DeepSeek-R1-Distilled-Qwen-1.5B using distributed reinforcement learning and a novel iterative context-lengthening strategy that gradually increases its context window from 8K to 24K tokens during training. It was trained on ~40,000 carefully curated mathematical problems drawn from competition-level datasets like AIME (1984–2023), AMC (pre-2023), Omni-MATH, and STILL. DeepScaleR achieves 43.1% accuracy on AIME 2024, a roughly 14.3 percentage point boost over the base model, and surpasses the performance of the proprietary O1-Preview model despite its much smaller size. It also posts strong results on a suite of math benchmarks (e.g., MATH-500, AMC 2023, Minerva Math, OlympiadBench), demonstrating that small, efficient models tuned with RL can match or exceed larger baselines on reasoning tasks.Starting Price: Free -
34
MiniMax M2.5
MiniMax
MiniMax M2.5 is a frontier AI model engineered for real-world productivity across coding, agentic workflows, search, and office tasks. Extensively trained with reinforcement learning in hundreds of thousands of real-world environments, it achieves state-of-the-art performance in benchmarks such as SWE-Bench Verified and BrowseComp. The model demonstrates strong architectural thinking, decomposing complex problems before generating code across more than ten programming languages. M2.5 operates at high throughput speeds of up to 100 tokens per second, enabling faster completion of multi-step tasks. It is optimized for efficient reasoning, reducing token usage and execution time compared to previous versions. With dramatically lower pricing than competing frontier models, it delivers powerful performance at minimal cost. Integrated into MiniMax Agent, M2.5 supports professional-grade office workflows, financial modeling, and autonomous task execution.Starting Price: Free -
35
Mistral 7B
Mistral AI
Mistral 7B is a 7.3-billion-parameter language model that outperforms larger models like Llama 2 13B across various benchmarks. It employs Grouped-Query Attention (GQA) for faster inference and Sliding Window Attention (SWA) to efficiently handle longer sequences. Released under the Apache 2.0 license, Mistral 7B is accessible for deployment across diverse platforms, including local environments and major cloud services. Additionally, a fine-tuned version, Mistral 7B Instruct, demonstrates enhanced performance in instruction-following tasks, surpassing models like Llama 2 13B Chat.Starting Price: Free -
36
Claude Sonnet 4.5
Anthropic
Claude Sonnet 4.5 is Anthropic’s latest frontier model, designed to excel in long-horizon coding, agentic workflows, and intensive computer use while maintaining safety and alignment. It achieves state-of-the-art performance on the SWE-bench Verified benchmark (for software engineering) and leads on OSWorld (a computer use benchmark), with the ability to sustain focus over 30 hours on complex, multi-step tasks. The model introduces improvements in tool handling, memory management, and context processing, enabling more sophisticated reasoning, better domain understanding (from finance and law to STEM), and deeper code comprehension. It supports context editing and memory tools to sustain long conversations or multi-agent tasks, and allows code execution and file creation within Claude apps. Sonnet 4.5 is deployed at AI Safety Level 3 (ASL-3), with classifiers protecting against inputs or outputs tied to risky domains, and includes mitigations against prompt injection. -
37
Molmo 2
Ai2
Molmo 2 is a new suite of state-of-the-art open vision-language models with fully open weights, training data, and training code that extends the original Molmo family’s grounded image understanding to video and multi-image inputs, enabling advanced video understanding, pointing, tracking, dense captioning, and question-answering capabilities; all with strong spatial and temporal reasoning across frames. Molmo 2 includes three variants: an 8 billion-parameter model optimized for overall video grounding and QA, a 4 billion-parameter version designed for efficiency, and a 7 billion-parameter Olmo-backed model offering a fully open end-to-end architecture including the underlying language model. These models outperform earlier Molmo versions on core benchmarks and set new open-model high-water marks for image and video understanding tasks, often competing with substantially larger proprietary systems while training on a fraction of the data used by comparable closed models. -
38
Kimi K2
Moonshot AI
Kimi K2 is a state-of-the-art open source large language model series built on a mixture-of-experts (MoE) architecture, featuring 1 trillion total parameters and 32 billion activated parameters for task-specific efficiency. Trained with the Muon optimizer on over 15.5 trillion tokens and stabilized by MuonClip’s attention-logit clamping, it delivers exceptional performance in frontier knowledge, reasoning, mathematics, coding, and general agentic workflows. Moonshot AI provides two variants, Kimi-K2-Base for research-level fine-tuning and Kimi-K2-Instruct pre-trained for immediate chat and tool-driven interactions, enabling both custom development and drop-in agentic capabilities. Benchmarks show it outperforms leading open source peers and rivals top proprietary models in coding tasks and complex task breakdowns, while its 128 K-token context length, tool-calling API compatibility, and support for industry-standard inference engines.Starting Price: Free -
39
Phi-2
Microsoft
We are now releasing Phi-2, a 2.7 billion-parameter language model that demonstrates outstanding reasoning and language understanding capabilities, showcasing state-of-the-art performance among base language models with less than 13 billion parameters. On complex benchmarks Phi-2 matches or outperforms models up to 25x larger, thanks to new innovations in model scaling and training data curation. With its compact size, Phi-2 is an ideal playground for researchers, including for exploration around mechanistic interpretability, safety improvements, or fine-tuning experimentation on a variety of tasks. We have made Phi-2 available in the Azure AI Studio model catalog to foster research and development on language models. -
40
Codestral
Mistral AI
We introduce Codestral, our first-ever code model. Codestral is an open-weight generative AI model explicitly designed for code generation tasks. It helps developers write and interact with code through a shared instruction and completion API endpoint. As it masters code and English, it can be used to design advanced AI applications for software developers. Codestral is trained on a diverse dataset of 80+ programming languages, including the most popular ones, such as Python, Java, C, C++, JavaScript, and Bash. It also performs well on more specific ones like Swift and Fortran. This broad language base ensures Codestral can assist developers in various coding environments and projects.Starting Price: Free -
41
SWE-1.6
Cognition
SWE-1.6 is an engineering–focused AI model developed by Cognition and integrated into the Windsurf environment, designed to optimize both raw intelligence and what the company calls “model UX,” or the overall feel and efficiency of interacting with an AI agent. It represents a new iteration in the SWE model family, improving performance on benchmarks such as SWE-Bench Pro by over 10% compared to SWE-1.5 while maintaining similar underlying capabilities. It was trained from scratch to jointly improve reasoning quality and user experience, addressing issues observed in earlier versions such as overthinking simple problems, taking too many steps, looping in repetitive reasoning, and relying excessively on terminal commands instead of specialized tools. SWE-1.6 introduces behavioral improvements such as more frequent parallel tool usage, faster context retrieval, and reduced need for user input, resulting in smoother and more efficient workflows. -
42
Olmo 2
Ai2
Olmo 2 is a family of fully open language models developed by the Allen Institute for AI (AI2), designed to provide researchers and developers with transparent access to training data, open-source code, reproducible training recipes, and comprehensive evaluations. These models are trained on up to 5 trillion tokens and are competitive with leading open-weight models like Llama 3.1 on English academic benchmarks. Olmo 2 emphasizes training stability, implementing techniques to prevent loss spikes during long training runs, and utilizes staged training interventions during late pretraining to address capability deficiencies. The models incorporate state-of-the-art post-training methodologies from AI2's Tülu 3, resulting in the creation of Olmo 2-Instruct models. An actionable evaluation framework, the Open Language Modeling Evaluation System (OLMES), was established to guide improvements through development stages, consisting of 20 evaluation benchmarks assessing core capabilities. -
43
Mistral Small
Mistral AI
On September 17, 2024, Mistral AI announced several key updates to enhance the accessibility and performance of their AI offerings. They introduced a free tier on "La Plateforme," their serverless platform for tuning and deploying Mistral models as API endpoints, enabling developers to experiment and prototype at no cost. Additionally, Mistral AI reduced prices across their entire model lineup, with significant cuts such as a 50% reduction for Mistral Nemo and an 80% decrease for Mistral Small and Codestral, making advanced AI more cost-effective for users. The company also unveiled Mistral Small v24.09, a 22-billion-parameter model offering a balance between performance and efficiency, suitable for tasks like translation, summarization, and sentiment analysis. Furthermore, they made Pixtral 12B, a vision-capable model with image understanding capabilities, freely available on "Le Chat," allowing users to analyze and caption images without compromising text-based performance.Starting Price: Free -
44
Mistral Large 2
Mistral AI
Mistral AI has launched the Mistral Large 2, an advanced AI model designed to excel in code generation, multilingual capabilities, and complex reasoning tasks. The model features a 128k context window, supporting dozens of languages including English, French, Spanish, and Arabic, as well as over 80 programming languages. Mistral Large 2 is tailored for high-throughput single-node inference, making it ideal for large-context applications. Its improved performance on benchmarks like MMLU and its enhanced code generation and reasoning abilities ensure accuracy and efficiency. The model also incorporates better function calling and retrieval, supporting complex business applications.Starting Price: Free -
45
GPT-5.4 mini
OpenAI
GPT-5.4 mini is a fast and efficient AI model designed for high-performance tasks such as coding, reasoning, and multimodal understanding. It delivers strong capabilities similar to larger models while maintaining lower latency and cost. The model is optimized for responsive applications where speed is critical, including coding assistants and real-time workflows. GPT-5.4 mini supports advanced features such as tool use, function calling, and image interpretation. It performs well on complex tasks while running significantly faster than previous mini models. The model is also suitable for subagent systems, where it handles smaller tasks within larger AI workflows. By combining speed, efficiency, and strong performance, GPT-5.4 mini enables scalable AI applications across various use cases. -
46
GPT-4.1 mini
OpenAI
GPT-4.1 mini is a compact version of OpenAI’s powerful GPT-4.1 model, designed to provide high performance while significantly reducing latency and cost. With a smaller size and optimized architecture, GPT-4.1 mini still delivers impressive results in tasks such as coding, instruction following, and long-context processing. It supports up to 1 million tokens of context, making it an efficient solution for applications that require fast responses without sacrificing accuracy or depth.Starting Price: $0.40 per 1M tokens (input) -
47
Mistral Large
Mistral AI
Mistral Large is Mistral AI's flagship language model, designed for advanced text generation and complex multilingual reasoning tasks, including text comprehension, transformation, and code generation. It supports English, French, Spanish, German, and Italian, offering a nuanced understanding of grammar and cultural contexts. With a 32,000-token context window, it can accurately recall information from extensive documents. The model's precise instruction-following and native function-calling capabilities facilitate application development and tech stack modernization. Mistral Large is accessible through Mistral's platform, Azure AI Studio, and Azure Machine Learning, and can be self-deployed for sensitive use cases. Benchmark evaluations indicate that Mistral Large achieves strong results, making it the world's second-ranked model generally available through an API, next to GPT-4.Starting Price: Free -
48
Leanstral
Mistral AI
Leanstral is an open-source code agent developed by Mistral AI specifically designed to work with the Lean 4 proof assistant. The model focuses on generating code while also formally verifying its correctness against strict mathematical or software specifications. Unlike traditional coding assistants, Leanstral integrates directly with formal proof systems to ensure that generated code satisfies defined logical requirements. Its architecture is optimized for proof engineering tasks and operates efficiently with sparse model parameters. Leanstral is released under the Apache 2.0 license, making it freely accessible for developers, researchers, and organizations to use and customize. The model is designed to operate within real-world formal repositories rather than isolated problem environments. By combining code generation with formal verification, Leanstral aims to reduce the need for manual human review in complex software and mathematical development.Starting Price: Free -
49
Mistral Saba
Mistral AI
Mistral Saba is a 24-billion-parameter model trained on meticulously curated datasets from across the Middle East and South Asia. The model provides more accurate and relevant responses than models that are over five times its size while being significantly faster and lower cost. It can also serve as a strong base to train highly specific regional adaptations. Mistral Saba is available as an API and can be deployed locally within customers' security premises. Like the recently released Mistral Small 3, the model is lightweight and can be deployed on single-GPU systems, responding at speeds of over 150 tokens per second. In keeping with the rich cultural cross-pollination between the Middle East and South Asia, Mistral Saba supports Arabic and many Indian-origin languages and is particularly strong in South Indian-origin languages such as Tamil. This capability enhances its versatility in multinational use across these interconnected regions.Starting Price: Free -
50
LTM-2-mini
Magic AI
LTM-2-mini is a 100M token context model: LTM-2-mini. 100M tokens equals ~10 million lines of code or ~750 novels. For each decoded token, LTM-2-mini’s sequence-dimension algorithm is roughly 1000x cheaper than the attention mechanism in Llama 3.1 405B1 for a 100M token context window. The contrast in memory requirements is even larger – running Llama 3.1 405B with a 100M token context requires 638 H100s per user just to store a single 100M token KV cache.2 In contrast, LTM requires a small fraction of a single H100’s HBM per user for the same context.
