Best cheap

GPT OSS 120B

Accounts

Open-weight GPT model for self-hosted reasoning and instruction-following workloads

GPT OSS 20B

Accounts

Open-weight GPT model for self-hosted reasoning and instruction-following workloads

DeepSeek V4 Flash

Accounts

Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work

MiniMax-M2.7

Accounts

MiniMax model for chat, coding, office work, and agentic tasks

Tongyi DeepResearch 30B A3B

Alibaba Nlp

General-purpose chat model for instruction following, writing, and analysis

Qwen3 235B A22B Instruct 2507

Alibaba

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen 3.32B

Alibaba

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen 3 Coder 30B A3B Instruct

Alibaba

Qwen coding model for software agents, repository edits, and code reasoning

Qwen3 Coder Next

Alibaba

Qwen coding model for software agents, repository edits, and code reasoning

Qwen3 Next 80B A3B Instruct

Alibaba

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen3 Next 80B A3B Thinking

Alibaba

Efficient Qwen thinking model for local reasoning, math, and coding agents

Nova Micro 1.0

Amazon

Amazon Nova Micro 1.0 is a text-only model that delivers the lowest latency responses in the Amazon Nova family of models at a very low cost. With a context length...

Nova Micro

Amazon

Efficient model for low-latency assistance, extraction, and routine automation

Flagship model for demanding analysis, coding, and production agent workflows

Trinity Large Thinking is a powerful open source reasoning model from the team at Arcee AI. It shows strong performance in PinchBench, agentic workloads, and reasoning tasks. Launc

Trinity Mini is a 26B-parameter (3B active) sparse mixture-of-experts language model featuring 128 experts with 8 active per token. Engineered for efficient reasoning over long con

Open-weight instruction model for adaptable chat and self-hosted production workloads

ERNIE-4.5-21B-A3B-Thinking

Baidu

Reasoning model for deliberate analysis, multi-step problem solving, and tool use

baidu/ERNIE-4.5-300B-A47B

Baidu

Tool-capable chat model for instruction following and agentic application workflows

Kimi K2 0711 Instruct FP4

Baseten

Kimi model for long-context chat, coding, and agentic reasoning

Tool-capable chat model for instruction following and agentic application workflows

Gemma Sea Lion V4 27B It

Cf

Open Gemma instruction model for efficient chat and self-hosted deployments

Granite 4.0 H Micro

Cf

Efficient model for low-latency assistance, extraction, and routine automation

Llama Guard 3 8B

Cf

Safety model for policy screening, moderation, and risk-aware routing workflows

Mistral Small 3.1 24B Instruct

Cf

Efficient Mistral model for fast chat, extraction, and production assistants

Nemotron 3 Super 120B

Cf

Nemotron middle tier for collaborative agents and high-volume reasoning workloads

GPT OSS 20B

Cf

Open-weight GPT model for self-hosted reasoning and instruction-following workloads

GPT OSS 120B

Cf

Open GPT reasoning model for self-hosted agents and controllable deployments

GLM-4.7-Flash

Cf

Efficient GLM model for fast reasoning, coding, and agent workflows

Mistral Small 3.2 24b Instruct

Chutesai

Efficient Mistral model for fast chat, extraction, and production assistants

Venice Uncensored Dolphin Mistral 24B Venice Edition is a fine-tuned variant of Mistral-Small-24B-Instruct-2501, developed by dphn.ai in collaboration with Venice.ai. This model is

Command R (08-2024)

Cohere

command-r-08-2024 is an update of the [Command R](/models/cohere/command-r) with improved performance for multilingual retrieval-augmented generation (RAG) and tool use. More broad

Cohere: Command R

Cohere

Cohere retrieval model for long-context chat and enterprise RAG workflows

Command R7B (12-2024)

Cohere

Command R7B (12-2024) is a small, fast update of the Command R+ model, delivered in December 2024. It excels at RAG, tool use, agents, and similar tasks requiring complex reasoning

DeepSeek-R1-0528

Deepseek Ai

DeepSeek reasoning model for multi-step analysis, math, coding, and tools

DeepSeek R1 Distill Llama 70B

Deepseek Ai

DeepSeek reasoning model for multi-step analysis, math, coding, and tools

deepseek-ai/DeepSeek-V3.1-Terminus

Deepseek Ai

DeepSeek chat model for instruction following, coding, and analysis

DeepSeek V3.1 Thinking

Deepseek Ai

DeepSeek chat model for instruction following, coding, and analysis

DeepSeek V3.1 Terminus (Thinking)

Deepseek Ai

DeepSeek chat model for instruction following, coding, and analysis

deepseek-ai/DeepSeek-V3.1

Deepseek Ai

DeepSeek chat model for instruction following, coding, and analysis

DeepSeek V3.2 Exp Thinking

Deepseek Ai

DeepSeek chat model for instruction following, coding, and analysis

deepseek-ai/DeepSeek-V3.2-Exp

Deepseek Ai

DeepSeek chat model for instruction following, coding, and analysis

DeepSeek-V3.2

Deepseek Ai

Legacy model retained for compatibility with older integrations

DeepSeek V4 Flash

Deepseek Ai

Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work

DeepSeek V4 Pro

Deepseek Ai

Open MoE flagship with million-token context for coding and long agent runs

DeepSeek V3 0324

DeepSeek

DeepSeek V3, a 685B-parameter, mixture-of-experts model, is the latest iteration of the flagship chat model family from the DeepSeek team. It succeeds the [DeepSeek V3](/deepseek/d

DeepSeek V3.1

DeepSeek

DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 ba

DeepSeek R1 0528 Qwen3 8B

DeepSeek

DeepSeek reasoning model for multi-step analysis, math, coding, and tools

R1 0528

DeepSeek

May 28th update to the [original DeepSeek R1](/deepseek/deepseek-r1) Performance on par with [OpenAI o1](/openai/o1), but open-sourced and with fully open reasoning tokens. It's 67

DeepSeek V3.1 Terminus

DeepSeek

DeepSeek-V3.1 Terminus is an update to [DeepSeek V3.1](/deepseek/deepseek-chat-v3.1) that maintains the model's original capabilities while addressing issues reported by users, inc

DeepSeek V3.2 Exp

DeepSeek

DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attentio

DeepSeek V4 Flash (Thinking)

DeepSeek

Fast DeepSeek model for efficient chat, coding help, and agent loops

DeepSeek V4 Flash

DeepSeek

DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window. I

DeepSeek V4 Pro Cheaper (Thinking)

DeepSeek

Flagship DeepSeek model for coding, reasoning, and agentic work

DeepSeek V4 Pro Cheaper

DeepSeek

Flagship DeepSeek model for coding, reasoning, and agentic work

DeepSeek V4 Pro

DeepSeek

DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window. It is designe

General-purpose chat model for instruction following, writing, and analysis

DiffusionGemma 26B-A4B IT

Google

Gemini model for general assistance, reasoning, and multimodal workflows

Gemini 1.5 Flash

Google

Fast Gemini model balancing multimodal reasoning, tool use, and cost

Nemotron 3 Super 120B A12B

Hf Nvidia

Nemotron middle tier for collaborative agents and high-volume reasoning workloads

GPT OSS 120B

Hf Openai

Open-weight GPT model for self-hosted reasoning and instruction-following workloads

GLM-4.7-Flash

Hf Zai Org

Budget GLM lane for fast coding help, routing, and everyday automation

Granite-4.0-H-Micro is a 3B parameter from the Granite 4 family of models. These models are the latest in a series of models released by IBM. They are fine-tuned for long...

Granite 4.1 8B is a dense, decoder-only 8-billion-parameter language model from IBM, part of the Granite 4.1 family. It supports a 131K-token context window and is designed for ent

Mercury 2 is an extremely fast reasoning LLM, and the first reasoning diffusion LLM (dLLM). Instead of generating tokens sequentially, Mercury 2 produces and refines multiple token

Ling-2.6-1T is an instant (instruct) model from inclusionAI and the company’s trillion-parameter flagship, designed for real-world agents that require fast execution and high effic

Ling-2.6-flash is an instant (instruct) model from inclusionAI with 104B total parameters and 7.4B active parameters, designed for real-world agents that require fast responses, st

Efficient model for low-latency assistance, extraction, and routine automation

Ring-2.6-1T is a 1T-parameter-scale thinking model with 63B active parameters, built for real-world agent workflows that require both strong capability and operational efficiency.

Mellum2-12B-A2.5B-Instruct is a fast MoE model with 131K context built for coding, tool use, and low-latency AI workflows.

Coding model for repository understanding, refactors, and agentic engineering tasks

Coding model for repository understanding, refactors, and agentic engineering tasks

Coding model for repository understanding, refactors, and agentic engineering tasks

KAT-Coder-Pro V2 is the latest high-performance model in KwaiKAT’s KAT-Coder series, designed for complex enterprise-grade software engineering and SaaS integration. It builds on t

Coding model for repository understanding, refactors, and agentic engineering tasks

Kimi reasoning model for long-horizon research, planning, and tool use

LongCat 2.0 is a sparse mixture-of-experts language model from Meituan, with 48B active parameters out of 1.6T total. It is suited for coding, repository-level changes, long-horizo

Llama 3.1 8B Instruct

Meta

Open Llama instruction model for multilingual chat, reasoning, and coding

Llama 3.2 1B Instruct

Meta

Open Llama instruction model for multilingual chat, reasoning, and coding

Llama 3.2 3B Instruct

Meta

Open Llama instruction model for multilingual chat, reasoning, and coding

Llama 3.1 70B Instruct

Meta

Meta's latest class of model (Llama 3.1) launched with a variety of sizes & flavors. This 70B instruct-tuned version is optimized for high quality dialogue usecases. It has demonst

Llama 3.1 8B Instruct

Meta

Meta's latest class of model (Llama 3.1) launched with a variety of sizes & flavors. This 8B instruct-tuned version is fast and efficient. It has demonstrated strong performance co

Llama 3.2 1B Instruct

Meta

Llama 3.2 1B is a 1-billion-parameter language model focused on efficiently performing natural language tasks, such as summarization, dialogue, and multilingual text analysis. Its

Llama 3.2 3B Instruct

Meta

Llama 3.2 3B is a 3-billion-parameter multilingual large language model, optimized for advanced natural language processing tasks like dialogue generation, reasoning, and summariza

Llama 3.3 70B Instruct

Meta

The Meta Llama 3.3 multilingual large language model (LLM) is a pretrained and instruction tuned generative model in 70B (text in/text out). The Llama 3.3 instruction tuned text on

Llama Guard 3 8B

Meta

Safety model for policy screening, moderation, and risk-aware routing workflows

MiniMax M2.1 Lightning

Minimax

High-speed MiniMax model for low-latency coding and agent workflows

MiniMax M2.1

Minimax

MiniMax-M2.1 is a lightweight, state-of-the-art large language model optimized for coding, agentic workflows, and modern application development. With only 10 billion activated par

MiniMax M2.5

Minimax

MiniMax-M2.5 is a SOTA large language model designed for real-world productivity. Trained in a diverse range of complex real-world digital working environments, M2.5 builds upon th

MiniMax M2.7

Minimax

MiniMax-M2.7 is a next-generation large language model designed for autonomous, real-world productivity and continuous improvement. Built to actively participate in its own evoluti

MiniMax M2

Minimax

MiniMax-M2 is a compact, high-efficiency large language model optimized for end-to-end coding and agentic workflows. With 10 billion activated parameters (230 billion total), it de

MiniMax-M2.5 High Throughput

Minimaxai

MiniMax model for chat, coding, office work, and agentic tasks

MiniMax-M2.1

Minimaxai

MiniMax model for chat, coding, office work, and agentic tasks

MiniMax M2.5 TEE

Minimaxai

Prior MiniMax coding model for agent workflows, office edits, and automation

MiniMax-M2.5

Minimaxai

Legacy model retained for compatibility with older integrations

MiniMax-M2.7

Minimaxai

MiniMax model for chat, coding, office work, and agentic tasks

MiniMax-M2

Minimaxai

Efficient open MiniMax model built for coding agents and tool-heavy workflows

Codestral (latest)

Mistral

Mistral code model for completions, refactors, and developer IDE workflows

Codestral (latest)

Mistral

Mistral code model for completions, refactors, and developer IDE workflows

Devstral 2

Mistral

Mistral's coding-agent model for repository work, terminal tasks, and software fixes

Devstral 2

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Devstral Medium

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Devstral 2 (latest)

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Devstral Small

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Devstral Small 1.1

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Magistral Small

Mistral

Mistral reasoning model for transparent analysis, math, and complex decisions

Ministral 3B (latest)

Mistral

Compact Mistral model for edge, latency-sensitive, and cost-efficient workloads

Ministral 8B (latest)

Mistral

Compact Mistral model for edge, latency-sensitive, and cost-efficient workloads

Mistral Nemo

Mistral

Mistral model for multilingual chat, reasoning, and tool-assisted workflows

Codestral 2508

Mistral

Mistral's cutting-edge language model for coding released end of July 2025. Codestral specializes in low-latency, high-frequency tasks such as fill-in-the-middle (FIM), code correc

Devstral 2 123B

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Devstral 2 2512

Mistral

Devstral 2 is a state-of-the-art open-source model by Mistral AI specializing in agentic coding. It is a 123B-parameter dense transformer model supporting a 256K context window. De

Mistral: Devstral Medium

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Devstral Small 2505

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Mistral: Devstral Small 1.1

Mistral

Mistral coding agent model for repository tasks and software engineering workflows

Mistral Nemo

Mistral

A 12B parameter model with a 128k token context length built by Mistral in collaboration with NVIDIA. The model is multilingual, supporting English, French, German, Spanish, Italia

Automatic model router for matching prompts to suitable backends and budgets

Automatic model router for matching prompts to suitable backends and budgets

DeepSeek V3.1 Nex N1

Nex Agi

DeepSeek chat model for instruction following, coding, and analysis

DeepHermes-3 Mistral 24B (Preview)

Nousresearch

Mistral model for multilingual chat, reasoning, and tool-assisted workflows

Hermes 4 Large (Thinking)

Nousresearch

Flagship model for demanding analysis, coding, and production agent workflows

Hermes 4 (Thinking)

Nousresearch

General-purpose chat model for instruction following, writing, and analysis

Hermes 4 70B

Nousresearch

Hermes 4 70B is a hybrid reasoning model from Nous Research, built on Meta-Llama-3.1-70B. It introduces the same hybrid mode as the larger 405B release, allowing the model to eithe

Llama 3.3 Nemotron Super 49B V1.5

NVIDIA

Llama-3.3-Nemotron-Super-49B-v1.5 is a 49B-parameter, English-centric reasoning/chat model derived from Meta’s Llama-3.3-70B-Instruct with a 128K context. It’s post-trained for age

Nvidia Nemotron Super 49B

NVIDIA

Nemotron model for efficient reasoning, coding, and specialized AI agents

Nemotron Super

NVIDIA

Nemotron middle tier for collaborative agents and high-volume reasoning workloads

Nemotron 3 Nano 30B A3B

NVIDIA

NVIDIA Nemotron 3 Nano 30B A3B is a small language MoE model with highest compute efficiency and accuracy for developers to build specialized agentic AI systems. The model is fully

NVIDIA Nemotron 3 Nano Omni

NVIDIA

Open Nemotron omni model combining reasoning with text, vision, and audio

Nvidia Nemotron 3 Super 120B Thinking

NVIDIA

Nemotron middle tier for collaborative agents and high-volume reasoning workloads

Nemotron 3 Super

NVIDIA

NVIDIA Nemotron 3 Super is a 120B-parameter open hybrid MoE model, activating just 12B parameters for maximum compute efficiency and accuracy in complex multi-agent applications. B

NVIDIA Nemotron Cascade 2

NVIDIA

Nemotron model for efficient reasoning, coding, and specialized AI agents

Nvidia Nemotron Nano 9B V2

NVIDIA

Compact Nemotron model for efficient reasoning and deployable AI agents

Nemotron 3 Super 120B A12B

NVIDIA

Nemotron middle tier for collaborative agents and high-volume reasoning workloads

Nemotron 3 Ultra 550B A55B

NVIDIA

Largest Nemotron 3 model for maximum open-weight reasoning and agent accuracy

GPT OSS 120B High Throughput

OpenAI

Open-weight GPT model for self-hosted reasoning and instruction-following workloads

GPT-4o-mini Search Preview

OpenAI

GPT-4o mini Search Preview is a specialized model for web search in Chat Completions. It is trained to understand and execute web search queries.

gpt-oss-120b

OpenAI

gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. I

gpt-oss-20b

OpenAI

gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per f

gpt-oss-safeguard-20b

OpenAI

gpt-oss-safeguard-20b is a safety reasoning model from OpenAI built upon gpt-oss-20b. This open-weight, 21B-parameter Mixture-of-Experts (MoE) model offers lower latency for safety

Laguna M.1 is the flagship coding agent model from [Poolside](https://poolside.ai/), optimized for complex software engineering tasks. Designed for agentic coding workflows, it sup

Laguna XS 2.1 is the latest coding agent model in the 33B-A3B category from [Poolside](https://poolside.ai/) and a step forward from their Laguna XS.2 model (released in April 2026

Reasoning model for deliberate analysis, multi-step problem solving, and tool use

Pro/deepseek-ai/DeepSeek-R1

Pro

DeepSeek reasoning model for multi-step analysis, math, coding, and tools

Legacy model retained for compatibility with older integrations

Pro/deepseek-ai/DeepSeek-V3.1-Terminus

Pro

Flagship DeepSeek model for coding, reasoning, and agentic work

Pro/deepseek-ai/DeepSeek-V3

Pro

Flagship DeepSeek model for coding, reasoning, and agentic work

Pro/deepseek-ai/DeepSeek-V3.2

Pro

Flagship DeepSeek model for coding, reasoning, and agentic work

Pro/MiniMaxAI/MiniMax-M2.5

Pro

Frontier MiniMax model for engineering, office tasks, and agentic reasoning

DeepSeek V3

Public

DeepSeek chat model for instruction following, coding, and analysis

MiniMax M2.5

Public

MiniMax model for chat, coding, office work, and agentic tasks

Qwen2.5 72B Instruct

Qwen

Qwen2.5 72B is the latest series of Qwen large language models. Qwen2.5 brings the following improvements upon Qwen2: - Significantly more knowledge and has greatly improved capabi

Qwen2.5 7B Instruct

Qwen

Qwen2.5 7B is the latest series of Qwen large language models. Qwen2.5 brings the following improvements upon Qwen2: - Significantly more knowledge and has greatly improved capabil

Qwen Plus 0728 (thinking)

Qwen

Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed, and cost combination.

Qwen Plus 0728

Qwen

Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed, and cost combination.

Qwen-Plus

Qwen

Qwen-Plus, based on the Qwen2.5 foundation model, is a 131K context model with a balanced performance, speed, and cost combination.

Qwen3 14B

Qwen

Qwen3-14B is a dense 14.8B parameter causal language model from the Qwen3 series, designed for both complex reasoning and efficient dialogue. It supports seamless switching between

Qwen3 235B A22B Instruct

Qwen

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen3 235B A22B Instruct 2507

Qwen

Qwen3-235B-A22B-Instruct-2507 is a multilingual, instruction-tuned mixture-of-experts language model based on the Qwen3-235B architecture, with 22B active parameters per forward pa

Qwen 3 235b A22B 2507 (TEE)

Qwen

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen3 235B A22B Instruct 2507 FP8

Qwen

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen 3 235b A22B 2507

Qwen

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen3 235B A22B Thinking 2507 TEE

Qwen

Qwen reasoning model for deliberate problem solving, math, and coding

Qwen3 235B A22B

Qwen

Qwen3-235B-A22B is a 235B parameter mixture-of-experts (MoE) model developed by Qwen, activating 22B parameters per forward pass. It supports seamless switching between a "thinking

Qwen3 235B A22B Thinking 2507

Qwen

Qwen3-235B-A22B-Thinking-2507 is a high-performance, open-weight Mixture-of-Experts (MoE) language model optimized for complex reasoning tasks. It activates 22B of its 235B paramet

Qwen3 30B A3B Instruct 2507

Qwen

Qwen3-30B-A3B-Instruct-2507 is a 30.5B-parameter mixture-of-experts language model from Qwen, with 3.3B active parameters per inference. It operates in non-thinking mode and is des

Qwen3 30B A3B Thinking 2507

Qwen

Qwen3-30B-A3B-Thinking-2507 is a 30B parameter Mixture-of-Experts reasoning model optimized for complex tasks requiring extended multi-step thinking. The model is designed specific

Qwen3 30B A3B

Qwen

Qwen3, the latest generation in the Qwen large language model series, features both dense and mixture-of-experts (MoE) architectures to excel in reasoning, multilingual support, an

Qwen3 32B

Qwen

Qwen3-32B is a dense 32.8B parameter causal language model from the Qwen3 series, optimized for both complex reasoning and efficient dialogue. It supports seamless switching betwee

Qwen3 4B

Qwen

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen3 4B Instruct

Qwen

Qwen instruction model for multilingual chat, reasoning, and tool use

Qwen3 8B

Qwen

Qwen3-8B is a dense 8.2B parameter causal language model from the Qwen3 series, designed for both reasoning-heavy tasks and efficient dialogue. It supports seamless switching betwe

Qwen3 Coder 30B A3B Instruct

Qwen

Qwen3-Coder-30B-A3B-Instruct is a 30.5B parameter Mixture-of-Experts (MoE) model with 128 experts (8 active per forward pass), designed for advanced code generation, repository-sca

Qwen3 Coder 480B A35B Instruct Turbo

Qwen

Qwen coding model for software agents, repository edits, and code reasoning

Qwen3 Coder 480B A35B Instruct

Qwen

Qwen coding model for software agents, repository edits, and code reasoning

Qwen3 Coder Flash

Qwen

Qwen3 Coder Flash is Alibaba's fast and cost efficient version of their proprietary Qwen3 Coder Plus. It is a powerful coding agent model specializing in autonomous programming via

Qwen3 Coder Next

Qwen

Qwen3-Coder-Next is an open-weight causal language model optimized for coding agents and local development workflows. It uses a sparse MoE design with 80B total parameters and only

Qwen3 Next 80B A3B Instruct

Qwen

Qwen3-Next-80B-A3B-Instruct is an instruction-tuned chat model in the Qwen3-Next series optimized for fast, stable responses without “thinking” traces. It targets complex tasks acr

Qwen3 Next 80B A3B Thinking

Qwen

Qwen3-Next-80B-A3B-Thinking is a reasoning-first chat model in the Qwen3-Next line that outputs structured “thinking” traces by default. It’s designed for hard multi-step problems;