Explore every AI model

206 AI models & tools for chat, images, video and audio.

Includes model variants and creative tools. Access varies by plan.

34 results

Claude

5

Claude Fable 5.1

[Core Function] Claude Fable 5.1 is Anthropic's premium model for demanding reasoning and long-horizon agentic work, succeeding Claude Fable 5. [Strengths] It brings stronger long-running agentic coding, multistep research, and document, spreadsheet, and slide work than Fable 5. Supports 1M-token context and 128K output. [Best For] Highly recommended for: the hardest reasoning problems, long autonomous agent runs, and high-stakes analytical work where capability matters more than cost. [Limitations] Do NOT use for routine chat; it is the slowest and most expensive Claude model (2.5x Opus 5.5). Non-default temperature, top_p, or top_k values return an error, and forced tool use is not supported. [Routing] Route here when the user asks for Fable, or when Claude Opus 5.5 at higher effort falls short.

ChatPro only

Claude Haiku 4.5

[Core Function] Claude Haiku 4.5 is Anthropic's fastest model, delivering near-frontier intelligence at a low cost. [Strengths] It matches Claude Sonnet 4's coding performance at one-third the cost and more than twice the speed, surpasses it on computer use, and scores 73.3% on SWE-bench Verified. [Best For] Highly recommended for: real-time chat assistants, customer service agents, pair programming, and parallel sub-agents. [Limitations] Do NOT use for very long documents or outputs: it has a 200K-token context and 64K output limit. It does not support the effort parameter. [Routing] Choose Haiku when speed and cost matter more than peak quality. Step up to Claude Sonnet 5 or Claude Opus 5.5 for richer reasoning.

ChatLite and above

Claude Opus 5

[Core Function] Claude Opus 5 is a legacy Opus model from Anthropic for coding and knowledge work, superseded by Claude Opus 5.5. [Strengths] At launch it came close to Claude Fable 5's intelligence at half the price, with strong results on software engineering, computer use, end-to-end business tasks, and scientific research. Supports 1M-token context and 128K output. [Best For] Highly recommended for: workflows already tuned and validated on Opus 5. [Limitations] Do NOT use for new work: Claude Opus 5.5 is newer, uses fewer tokens per task, and costs 20% less per token. [Routing] Route here only if the user explicitly requests Opus 5. Otherwise, use Claude Opus 5.5.

ChatPro only

Claude Opus 5.5

[Core Function] Claude Opus 5.5 is Anthropic's recommended default model for most workloads, built for long-running agentic coding and knowledge work. [Strengths] It leads Anthropic's benchmarks in agentic coding, computer use, and knowledge work, excels at codebase-wide migrations and audits, reads dense charts and screenshots precisely, and uses fewer tokens per task than Opus 5. Supports 1M-token context and 128K output. [Best For] Highly recommended for: coding, research and analysis, document understanding, and agent workflows. [Limitations] Do NOT use for simple, high-volume tasks; use Claude Haiku 4.5. Thinking cannot be disabled, and forced tool use is not supported. [Routing] Use this model by default for chat requests. Route to Claude Fable 5.1 for the most demanding reasoning or long-horizon agentic work.

ChatPro only

Claude Sonnet 5

[Core Function] Claude Sonnet 5 is a legacy Sonnet-class model from Anthropic, offering performance close to Claude Opus 4.8 at a lower price. [Strengths] It is the most agentic Sonnet model to date, able to plan, use tools such as browsers and terminals, and run autonomously, with selectable effort levels. Supports 1M-token context and 128K output. [Best For] Highly recommended for: everyday coding, tool use, and knowledge work where cost matters. [Limitations] Do NOT use for the most demanding reasoning. Non-default temperature, top_p, or top_k values return an error, and cyber safeguards block dangerous security usage. [Routing] Route here for cost-sensitive everyday tasks that need more than Claude Haiku 4.5. Use Claude Opus 5.5 when quality matters.

ChatPlus and above

DeepSeek

4

Deepseek V4 Flash

[Core Function] DeepSeek V4 Flash was DeepSeek's fast, economical V4-generation model. DeepSeek has retired it, and requests to it are now served by DeepSeek V4.1 Flash. [Strengths] Responses come from V4.1 Flash, which DeepSeek reports is smarter, faster, and cheaper, with 1M-token context, up to 384K output, image input, and tool calls. [Best For] Highly recommended for: existing integrations that still reference V4 Flash. [Limitations] Do NOT expect original V4 Flash behaviour; the model name is a temporary compatibility alias and may stop working. [Routing] Route new requests to DeepSeek V4.1 Flash directly.

ChatFree trial included

Deepseek V4 Flash Vision

[Core Function] DeepSeek V4 Flash Vision was DeepSeek's experimental multimodal model. DeepSeek has retired it, and requests to it are now served by DeepSeek V4.1 Flash, which has native visual understanding. [Strengths] Responses come from V4.1 Flash, which supports mixed text and image input, 1M-token context, up to 384K output, and tool calls. [Best For] Highly recommended for: existing integrations that still reference V4 Flash Vision. [Limitations] Do NOT expect original V4 Flash Vision behaviour; the model name is a temporary compatibility alias and may stop working. [Routing] Route new image-understanding requests to DeepSeek V4.1 Flash directly.

ChatLite and above

Deepseek V4 Pro

[Core Function] DeepSeek V4 Pro is DeepSeek's large flagship model from the V4 generation, with 1.6T total and 49B active parameters. [Strengths] At release it led open-source models on agentic coding and math, STEM, and coding reasoning, with major agent upgrades in its August 2026 update. Supports 1M-token context, up to 384K output, tool calls, JSON output, and low, high, or max reasoning effort. [Best For] Highly recommended for: workflows already tuned and validated on DeepSeek V4 Pro. [Limitations] Do NOT use for new work: DeepSeek reports V4.1 Flash beats it on performance, cost, and speed. It does not support image input, and temperature is not supported in thinking mode. [Routing] Route here only if the user explicitly asks for V4 Pro. Otherwise, use DeepSeek V4.1 Flash.

ChatLite and above

Deepseek V4.1 Flash

[Core Function] DeepSeek V4.1 Flash is DeepSeek's latest model and the smallest in its new architecture family, with native visual understanding. [Strengths] It is a 552B-parameter MoE model that DeepSeek reports outperforms DeepSeek V4 Pro on benchmarks, cost, and speed. Supports 1M-token context, up to 384K output, text and image input, tool calls, JSON output, and low, high, or max reasoning effort. [Best For] Highly recommended for: coding agents, reasoning, image understanding, and cost-sensitive high-volume workloads. [Limitations] In thinking mode (the default), temperature, presence_penalty, and frequency_penalty are not supported. [Routing] Use this model by default for all DeepSeek requests, including image inputs. Route to DeepSeek V4 Pro only if the user explicitly asks for it.

ChatPro only

Gemini

4

Gemini 3.1 Pro

[Core Function] Gemini 3.1 Pro is Google's third-generation Pro model, built for multimodal understanding, agentic capabilities, and vibe coding. [Strengths] It offers better thinking, improved token efficiency, and more grounded, factually consistent output than Gemini 3 Pro, and is optimised for software engineering and precise multi-step tool use. Supports multimodal input, 1M-token context, and 64K output. [Best For] Highly recommended for: complex problem-solving, agentic coding, and multimodal understanding. [Limitations] It is still a Preview model, last updated February 2026, and costs more per token than Gemini 3.8 Flash ($2/$12 vs an introductory $0.75/$3.75), more for prompts above 200K tokens. [Routing] Route here when the user explicitly asks for Gemini Pro. Otherwise, use Gemini 3.8 Flash.

ChatPlus and above

Gemini 3.6 Flash

[Core Function] Gemini 3.6 Flash is an earlier-generation Google Flash model that balances speed and multimodal capabilities across general agentic and everyday tasks. [Strengths] It excels at code generation, agentic execution, and spatial reasoning, and is particularly effective for rapid agentic loops with repeated coding iterations. Supports multimodal input, 1M-token context, and 64K output. [Best For] Highly recommended for: workflows already tuned on Gemini 3.6 Flash, and fast iterative coding loops. [Limitations] Do NOT use for new work: Gemini 3.7 Flash and Gemini 3.8 Flash are newer and cost the same per token. [Routing] Route here only if the user explicitly asks for Gemini 3.6 Flash. Otherwise, use Gemini 3.8 Flash, or Gemini 3.7 Flash for lighter tasks.

ChatLite and above

Gemini 3.7 Flash

[Core Function] Gemini 3.7 Flash is Google's previous-generation Flash model for complex coding, agentic workflows, and reliable multi-step execution. [Strengths] It is a high-speed, efficient model built for everyday coding and agentic tool use, with multimodal input. Supports 1M-token context, 64K output, and low, medium, or high thinking levels. [Best For] Highly recommended for: everyday coding, tool-using agents, and routine multi-step tasks where token use matters. [Limitations] Do NOT use for the most demanding long-horizon work; Gemini 3.8 Flash is more capable at the same price. The minimal thinking level returns an error. [Routing] Route here for everyday tasks where Gemini 3.8 Flash would use more tokens than needed, or when the user asks for 3.7 Flash. Use Gemini 3.8 Flash for complex work.

ChatLite and above

Gemini 3.8 Flash

[Core Function] Gemini 3.8 Flash is Google's most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows. [Strengths] It delivers strong results on complex multi-file refactoring and tool execution, reduces failed loops in multi-step agents, and accepts text, image, video, audio, and PDF input. Supports 1M-token context, 64K output, and low, medium, or high thinking levels. [Best For] Highly recommended for: complex coding, agent workflows, and demanding enterprise tasks. [Limitations] It can use more tokens on long, complex tasks by design. The minimal thinking level returns an error, and temperature, top_p, and top_k should not be set. [Routing] Use this model by default for Gemini requests. Route to Gemini 3.7 Flash for lighter everyday tasks where token use matters.

ChatLite and above

GLM

4

GLM 4.7 Flash

[Core Function] GLM 4.7 Flash is Z.ai's free, lightweight model from the GLM-4.7 series. [Strengths] The GLM-4.7 series brings enhanced programming capabilities, more stable multi-step reasoning and execution, more natural conversation, and better front-end aesthetics. Supports English and Chinese with a 200K-token context window, at no cost. [Best For] Highly recommended for: simple chat, lightweight coding help, and high-volume tasks where cost must be zero. [Limitations] Do NOT use for complex coding, long-horizon agents, or image input; it is an older, lightweight model. [Routing] Choose this model when a free option is required. Step up to GLM 5.3 Flash for stronger, multimodal, low-cost performance, or GLM 5.3 for complex work.

ChatFree trial included

GLM 5.2

[Core Function] GLM 5.2 is an earlier Z.ai flagship model built for long-horizon tasks, with a fully usable 1M-token context window. [Strengths] It handles project-scale engineering context, executes long tasks more stably, adheres more reliably to engineering standards, and supports client-side and mobile engineering with on-device debugging. Supports text input, 128K output, thinking modes, function calling, context caching, structured output, and MCP tools. [Best For] Highly recommended for: workflows already tuned and validated on GLM 5.2. [Limitations] Do NOT use for new work: GLM 5.3 is newer and stronger at the same price, and GLM 5.3 Flash outperforms it at much lower cost. [Routing] Route here only if the user explicitly asks for GLM 5.2. Otherwise, use GLM 5.3.

ChatPlus and above

GLM 5.3

[Core Function] GLM 5.3 is Z.ai's latest flagship model, built for complex software engineering and agent tasks. [Strengths] It improves coding performance by 50% over GLM 5.2 on Z.ai's in-house Code Bench while using fewer output tokens, reaches state-of-the-art results among open-source models on Terminal Bench 3.0, and shows strong vulnerability discovery capability. Supports 1M-token context, 128K output, function calling, structured output, and low, high, or max reasoning effort. [Best For] Highly recommended for: complex coding, long-horizon agent tasks, terminal work, and security analysis. [Limitations] Text input only; it cannot read images or video. Reasoning is always on and cannot be disabled. [Routing] Use this model by default for text-based GLM requests. Route to GLM 5.3 Flash for image or video input, or when cost matters.

ChatLite and above

GLM 5.3 Flash

[Core Function] GLM 5.3 Flash is the first native multimodal model in Z.ai's GLM-5 series, delivering stronger intelligence than GLM 5.2 at very low cost. [Strengths] Vision is built into its coding loop: it inspects interfaces, rendered output, and interaction feedback to improve its work across frontend, game, and 3D development. It also handles office tasks, financial research, and PPTX, PDF, DOCX, and XLSX deliverables. Supports 1M-token context with image, video, and file input. [Best For] Highly recommended for: visual coding, document work, and cost-sensitive workloads. [Limitations] Thinking cannot be disabled. [Routing] Choose this model for image, video, or file input, or when cost matters. Step up to GLM 5.3 for the hardest coding tasks.

ChatFree trial included

GPT

7

GPT 5.5

[Core Function] GPT-5.5 is an earlier OpenAI flagship model for complex coding and professional work. [Strengths] It is a strong fit for coding, tool-heavy agents, grounded assistants, long-context retrieval, and customer-facing workflows where response polish matters, reaching strong results with fewer reasoning tokens than prior models. Supports 1.05M-token context and 128K output. [Best For] Highly recommended for: workflows already tuned and validated on GPT-5.5. [Limitations] Do NOT use for new work: it costs more per token ($5/$30) than GPT-5.6 Sol ($4/$20) and GPT-6 Sol ($2/$10). It does not support the max reasoning effort. [Routing] Route here only if the user explicitly asks for GPT-5.5. Otherwise, use GPT-6 Astra.

ChatPro only

GPT 5.6 Luna

[Core Function] GPT-5.6 Luna is OpenAI's previous-generation GPT-5.6 model for cost-sensitive, high-volume workloads, equivalent to the earlier nano tier. [Strengths] It is the lowest-cost GPT-5.6 model ($0.20/$1.20 per million tokens), with 1.05M-token context, 128K output, and six reasoning effort levels (none to max). [Best For] Highly recommended for: workflows already tuned on GPT-5.6 Luna, with cost-sensitive, high-volume workloads. [Limitations] Do NOT use for new work: GPT-6 Luna is newer and costs half as much per token ($0.10/$0.50). [Routing] Route here only if the user explicitly asks for GPT-5.6 Luna. Otherwise, use GPT-6 Luna.

ChatFree trial included

GPT 5.6 Sol

[Core Function] GPT-5.6 Sol is the flagship model of OpenAI's previous-generation GPT-5.6 family, built for complex professional work. [Strengths] GPT-5.6 is especially token-efficient and improves frontend design, including layout, visual hierarchy, and design judgment. Supports 1.05M-token context, 128K output, and six reasoning effort levels (none to max). [Best For] Highly recommended for: workflows already tuned on GPT-5.6, and polished website and app frontends. [Limitations] Do NOT use for new work: GPT-6 Sol costs half as much per token ($2/$10 vs $4/$20), and GPT-6 Astra is more capable. [Routing] Route here only if the user explicitly asks for GPT-5.6 Sol. Otherwise, use GPT-6 Astra for demanding work or GPT-6 Sol for lower cost.

ChatPro only

GPT 5.6 Terra

[Core Function] GPT-5.6 Terra is OpenAI's previous-generation GPT-5.6 model that balances intelligence and cost, equivalent to the earlier mini tier. [Strengths] It shares GPT-5.6's token efficiency and improved frontend design at a lower price than GPT-5.6 Sol ($2/$12 vs $4/$20). Supports 1.05M-token context, 128K output, and six reasoning effort levels (none to max). [Best For] Highly recommended for: workflows already tuned on GPT-5.6 Terra that need solid performance at moderate cost. [Limitations] Do NOT use for new work: GPT-6 Sol offers OpenAI's highest reasoning tier at a similar price ($2/$10 vs $2/$12), and GPT-6 Luna is far cheaper for simple tasks. [Routing] Route here only if the user explicitly asks for GPT-5.6 Terra. Otherwise, use GPT-6 Sol or GPT-6 Luna.

ChatPlus and above

GPT 6 Astra

[Core Function] GPT-6 Astra is OpenAI's most capable model and its recommended starting point for complex reasoning and coding. [Strengths] It delivers state-of-the-art performance in computer use, browsing, software engineering, science, and professional work, using substantially fewer output tokens than earlier models in several evaluations. Supports 1.05M-token context and 128K output. [Best For] Highly recommended for: ambiguous problems, deep analysis, complex coding, research, and ambitious deliverables. [Limitations] Do NOT use for simple, high-volume tasks; it costs $10/$50 per million tokens. The none reasoning effort is not supported, so temperature and top_p cannot be set, and tool calling requires the Responses API. [Routing] Use this model by default for demanding OpenAI requests. Route to GPT-6 Sol to reduce cost, or GPT-6 Luna for high-volume tasks.

ChatPro only

GPT 6 Luna

[Core Function] GPT-6 Luna is OpenAI's most efficient model for focused, high-volume tasks. [Strengths] It is OpenAI's fastest and most cost-effective GPT-6 model ($0.10/$0.50 per million tokens), with 1.05M-token context, 128K output, and six reasoning effort levels (none to max). [Best For] Highly recommended for: scoped tasks, triage, frequent automations, fine-grained edits, and simple data extraction. [Limitations] Do NOT use for ambiguous problems or deep analysis; use GPT-6 Astra. In Chat Completions, function calling works only with reasoning effort set to none, and temperature and top_p are only supported at none. [Routing] Choose Luna when cost and speed matter more than peak quality. Step up to GPT-6 Sol for complex work, or GPT-6 Astra for the most demanding tasks.

ChatPro only

GPT 6 Sol

[Core Function] GPT-6 Sol is OpenAI's GPT-6 model built for complex coding and agentic workflows, balancing intelligence and cost. [Strengths] It offers OpenAI's highest reasoning tier at one-fifth of GPT-6 Astra's price, with 1.05M-token context, 128K output, and six reasoning effort levels (none to max). [Best For] Highly recommended for: complex coding, agent workflows, and professional work where cost matters. [Limitations] OpenAI has released GPT-6.1 Sol as its newer Sol model. In Chat Completions, function calling works only with reasoning effort set to none, and temperature and top_p are only supported at none. [Routing] Route here for complex tasks where cost matters more than peak quality. Step up to GPT-6 Astra for the most demanding work, or down to GPT-6 Luna for high-volume tasks.

ChatPro only

Grok

3

Grok 4.5

[Core Function] Grok 4.5 is an earlier xAI coding model for agentic software, engineering, and workflow tasks, trained on data spanning coding, science, engineering, and math. [Strengths] It is capable at Rust and C/C++ tasks and end-to-end app building from minimal specifications, and is served at about 80 tokens per second. Supports 500K-token context, text and image input, function calling, structured outputs, and four reasoning effort levels. [Best For] Highly recommended for: workflows already tuned and validated on Grok 4.5. [Limitations] Do NOT use for new work: Grok 4.7 and Grok 4.6 are newer at the same input and output price. [Routing] Route here only if the user explicitly asks for Grok 4.5. Otherwise, use Grok 4.7.

ChatPlus and above

Grok 4.6

[Core Function] Grok 4.6 is an earlier xAI model for coding, agentic tasks, and knowledge work, focused on long-running agents and ambitious interactive and visual work. [Strengths] It is especially strong at turning a broad product idea into a working first version, producing stronger first passes on visual and interactive projects than Grok 4.5, with more self-testing on long tasks. Supports 500K-token context, text and image input, function calling, structured outputs, and four reasoning effort levels. [Best For] Highly recommended for: workflows already tuned and validated on Grok 4.6. [Limitations] Do NOT use for new work: Grok 4.7 is more capable at the same price and speed. [Routing] Route here only if the user explicitly asks for Grok 4.6. Otherwise, use Grok 4.7.

ChatPlus and above

Grok 4.7

[Core Function] Grok 4.7 is xAI's most capable model, built for coding, agentic tasks, and knowledge work. [Strengths] Built on a new, larger base model than Grok 4.6, it works longer on difficult tasks, checks its own work more carefully, manages long context better, and is stronger at creating documents and presentations. Supports 500K-token context, text and image input, function calling, structured outputs, web and X search, and four reasoning effort levels. [Best For] Highly recommended for: long-running coding, professional knowledge work, and agent workflows. [Limitations] Prompts of 200K tokens or more are billed at twice the standard rate. It has no knowledge of current events unless search tools are enabled. [Routing] Use this model by default for all Grok requests.

ChatPro only

Kimi

2

Kimi K2.7

[Core Function] Kimi K2.7 (Kimi K2.7 Code) is Moonshot AI's dedicated coding model. [Strengths] It follows instructions more reliably in long contexts and completes coding tasks with higher success rates than Kimi K2.6, with about 10% better agentic performance and 30% less overthinking on average. Supports 256K-token context, visual input, and multi-step tool calling. [Best For] Highly recommended for: cost-sensitive coding, long-context code tasks, and coding agents. [Limitations] Do NOT use for broad knowledge work or the hardest reasoning; use Kimi K3. Thinking cannot be disabled, temperature and top_p are fixed, and tool_choice "required" is not supported. [Routing] Route here for everyday coding where cost matters (about one-third of K3's per-token price). Step up to Kimi K3 for long-horizon or complex work.

ChatLite and above

Kimi K3

[Core Function] Kimi K3 is Moonshot AI's most capable flagship model to date, a 2.8-trillion-parameter open-source model built for long-horizon coding, knowledge work, and deep reasoning. [Strengths] It sustains long-running engineering tasks with minimal supervision, works across large codebases, and uses screenshots and visual feedback for frontend, game, and CAD work. Supports 1M-token context, native image and video understanding, and low, high, or max reasoning effort. [Best For] Highly recommended for: complex coding agents, knowledge work, and visual coding tasks. [Limitations] Thinking is always on and defaults to max effort; lower it for simple requests to save time and tokens. Temperature and top_p are fixed and other values return an error. [Routing] Use this model by default for Kimi requests. Route to Kimi K2.7 for lower-cost coding tasks.

ChatPro only

Modellix

1

Free LLM

Open-source LLM

ChatFree trial included

Qwen

4

Qwen 3.7 Max

[Core Function] Qwen 3.7 Max is a legacy Max-tier Qwen model that Alibaba Cloud no longer recommends. [Strengths] It supports thinking and non-thinking modes, function calling, built-in tools, structured output, and a 1M-token context window. [Best For] Highly recommended for: workflows already tuned and validated on Qwen 3.7 Max. [Limitations] Do NOT use for new work: Alibaba Cloud lists it as legacy and no longer recommended. [Routing] Route here only if the user explicitly asks for Qwen 3.7 Max. Otherwise, use Qwen 3.8 Max.

ChatLite and above

Qwen 3.7 Plus

[Core Function] Qwen 3.7 Plus is Alibaba Cloud's recommended starting model for most text tasks, balancing performance and cost. [Strengths] It offers full function calling, built-in tools such as web search and code interpreter, structured output, and hybrid thinking mode, with a 1M-token context window suited to large codebases and long documents. It also supports image and video understanding. [Best For] Highly recommended for: chatbots, content generation, summarisation, document processing, office tasks, and coding tools. [Limitations] Do NOT use when the strongest reasoning is required; use Qwen 3.8 Max. Prompts above 256K tokens are billed at three times the standard rate. [Routing] Use this model by default for Qwen requests. Route to Qwen 3.8 Max for the hardest reasoning, or Qwen 3.8 Flash to cut costs.

ChatFree trial included

Qwen 3.8 Flash

[Core Function] Qwen 3.8 Flash is Alibaba Cloud's lightweight, low-cost Qwen model. [Strengths] Alibaba describes it as delivering near-flagship performance at a lower price, with thinking mode, function calling, built-in tools, structured output, and a 1M-token context window. [Best For] Highly recommended for: cost-sensitive chat, summarisation, office tasks, and high-volume workloads. [Limitations] Do NOT use for the hardest reasoning tasks; use Qwen 3.8 Max. [Routing] Choose Flash when cost matters more than peak quality. Step up to Qwen 3.7 Plus for balanced performance, or Qwen 3.8 Max for the strongest reasoning.

ChatLite and above

Qwen 3.8 Max

[Core Function] Qwen 3.8 Max is Alibaba Cloud's highest-capability Qwen model, built for the strongest reasoning. [Strengths] It supports thinking and non-thinking modes, function calling, built-in tools, structured output, image and video understanding, and a 1M-token context window. [Best For] Highly recommended for: complex reasoning, difficult coding, multi-step analysis, and tasks where quality matters more than cost. [Limitations] Do NOT use for routine chat or high-volume tasks; it costs $2/$6 per million tokens. [Routing] Route here when a request needs Qwen's strongest reasoning. Use Qwen 3.7 Plus for everyday tasks.

ChatPro only