Compare pricing and specifications for large language models from all major providers.
Anthropic's flagship Opus-class model for July 2026. Opus 5 pairs Mythos-class reasoning with the Opus price point, adds adaptive effort control and a hardened long-horizon agent loop that holds context across day-long tasks.
The 1M-token context tier of Claude Opus 5. Same model, long-context premium pricing applied to requests above 200K input tokens. Built for whole-repository and whole-corpus work.
The workhorse of the Claude 5 family. Sonnet 5 lands close to Opus 4.8 on coding and agentic benchmarks while staying at the long-standing Sonnet price, with a 1M token context window available by default.
OpenAI's Sol tier for GPT-5.6: a long-thinking configuration that spends far more reasoning tokens per request and tops OpenAI's own agentic and math evaluations. Priced as a premium reasoning endpoint.
The July 2026 refresh of the GPT-5 series. GPT-5.6 keeps GPT-5.5 pricing while improving instruction following, tool-call accuracy and refusal calibration, and it ships with a faster streaming stack.
The cost tier of GPT-5.6. Mini keeps the same tokenizer and tool interface as the full model, trading depth for roughly an eighth of the price and noticeably lower latency.
The Codex-tuned build of GPT-5.6 for the Codex CLI, IDE extension and cloud agents. Optimised for long autonomous coding sessions, patch generation and test repair rather than open-ended chat.
Google's flagship Gemini for mid-2026. Gemini 3.6 Pro extends the 2M token window to all paid tiers, improves grounded answers with live Search, and adds native 1 fps video understanding.
The fast, cheap tier of Gemini 3.6. Flash keeps a 1M token window and multimodal input while running several times faster than Pro, which makes it the default choice for high-volume production traffic.
Moonshot's flagship Kimi K3.7, an open-weight mixture-of-experts model with a 512K token window. K3.7 closes most of the gap to Western frontier models on agentic coding while staying dramatically cheaper.
The low-latency hosted tier of Kimi K3.7. Same weights, dedicated high-throughput serving, roughly double the price and a large jump in tokens per second.
A distilled K3.7 aimed at classification, extraction and sub-agent duty. Runs comfortably on a single high-memory GPU when self-hosted.
Alibaba's Qwen 3.8 flagship. A sparse MoE model with 256K context, first-class support for more than 100 languages and a notable jump in tool-calling reliability over Qwen 3.
The coding-specialised Qwen 3.8 with a 1M token window. Tuned for repository-scale understanding, patch generation and terminal agents.
Anthropic's most powerful frontier model - the Mythos-class system with state-of-the-art agentic and cybersecurity capability. Distributed under strict access controls; Fable 5 is its safeguarded public sibling.
The first public Mythos-class model - Anthropic's most powerful generally available model, sharing Mythos 5's underlying intelligence with added safeguards that fall back to Opus 4.8 on sensitive prompts.
Meta's flagship open-weight model for 2026. Llama 5 Maverick is a large mixture-of-experts system with a 2M token window and native multimodal input, released under the Llama 5 Community License.
The compact member of the Llama 5 family. Scout runs on a single 24 GB consumer GPU at full speed while keeping a 1M token window and vision input, which makes it the practical choice for local and edge work.
Meta's largest released model, positioned against closed frontier systems. Behemoth is served through partners rather than being practical to self-host, and it leads open-weight benchmarks on reasoning and math.
A mid-cycle refresh of DeepSeek V4 with a doubled context window, better tool calling and a large improvement in multilingual output quality. Still open weights.
xAI's Grok 4 with a 1M token window, live X and web grounding, and a large jump in coding benchmarks over Grok 3.5. Available through the xAI API and X Premium+.
The cost tier of Grok 4. Keeps live grounding and the 1M window but targets high-volume traffic at roughly a tenth of the flagship price.
Anthropic's most capable Opus-tier model - highly autonomous, state-of-the-art on long-horizon agentic work, knowledge work and memory, with clearer, warmer writing. 1M context at standard pricing.
Google's newest fast model (Google I/O 2026), built for agentic tasks and coding - beats Gemini 3.1 Pro on coding and agentic benchmarks at a fraction of the cost.
Mistral's refreshed flagship with a 256K window, stricter JSON mode and EU data residency through La Plateforme. The default pick for European teams with GDPR locality requirements.
Zhipu's GLM-5, an open-weight MoE model with strong bilingual Chinese and English performance and a competitive agentic coding score at a fraction of frontier pricing.
OpenAI's flagship GPT-5.5, released April 2026. A ~1.05M token context window with strong terminal/agentic benchmarks; available in the Responses and Chat Completions APIs.
A higher-accuracy GPT-5.5 variant for the most demanding tasks, available via the API at premium pricing.
OpenAI's most advanced model with enhanced reasoning, longer context, and multimodal capabilities. Top of the line for complex tasks.
Affordable version of GPT-5.4 with strong performance for everyday tasks at a fraction of the cost.
Fast and affordable Gemini 3.1 Flash optimized for high-throughput applications at minimal cost.
xAI's latest flagship model with expanded context, improved reasoning, and deeper integration with real-time data from X platform.
Previous-generation Opus - highly autonomous with excellent long-horizon agentic work, vision and memory. Adaptive thinking only; high-resolution vision support.
DeepSeek's second-generation reasoning model. R2 exposes its full chain of thought, scores near the top of competitive math benchmarks and is priced far below comparable Western reasoning endpoints.
Anthropic's Claude Opus 4.6 with extended 1M token context window for processing entire codebases, books, and massive datasets in a single prompt.
Anthropic's latest flagship model with best-in-class reasoning, coding, and agentic capabilities. Supports extended thinking for complex multi-step problems.
Anthropic's balanced mid-tier model offering strong intelligence, speed, and cost-effectiveness. Excellent for everyday coding and enterprise tasks.
Google's most capable Gemini Pro model with an industry-leading 2M token context window - ideal for large-document analysis and long multi-turn reasoning.
DeepSeek's fourth-generation open-weights model with state-of-the-art reasoning at remarkably low cost. Strong performance on coding and math benchmarks.
An o3-family model optimized for deep research tasks. Autonomously browses the web, synthesizes information, and produces comprehensive research reports.
Mistral's latest flagship model with multilingual excellence, strong coding, and enterprise-grade function calling. Open-weight model.
Alibaba's latest Qwen 3 flagship model with strong multilingual capabilities and improved reasoning. Competitive with frontier models at lower cost.
Professional AI image generation with photorealistic quality and artistic control. Subscription-based: Basic $10/mo, Standard $30/mo, Pro $60/mo
Higher-compute variant of GPT-5.2 for harder problems (Responses API only).
Most intelligent Codex model optimized for long-horizon agentic coding (Responses API only).
Gemini 3 Flash Preview on Vertex AI.
OpenAI flagship model for coding and agentic tasks across industries.
Flagship GPT model with configurable reasoning effort; predecessor to GPT-5.2.
GPT-5.1 variant optimized for agentic coding in Codex (Responses API only).
Claude 4.5 flagship Opus model for long-horizon coding and agentic workflows.
Gemini 3 Pro Preview on Vertex AI.
Anthropic's fastest 4.5-series model-cost-effective for high-volume workloads with extended thinking and strong tool use.
Windsurf/Cognition in-house frontier model for agentic coding. Consumed via Windsurf prompt credits (not USD per-token).
Anthropic's fastest and most cost-effective 4.5-series model, optimized for high-volume workloads with strong tool use capabilities.
A version of GPT-5 optimized for agentic coding in Codex. Default in Codex cloud & reviews; also usable via API key. Priced the same as GPT-5.
Anthropic's most intelligent model for agents and coding, with extended thinking and state-of-the-art performance on SWE-bench Verified.
Next-generation Gemini with advanced multimodal reasoning and long context. Rates pending official pricing page.
Opus 4.1 (200k) - higher-cost flagship Opus tier. Newer Opus 4.5 provides a more accessible price point.
Version of GPT-5 that produces smarter and more precise responses. Responses API only; higher max output than standard GPT-5.
Previous flagship GPT model for coding, reasoning, and agentic tasks. OpenAI recommends GPT-5.1/5.2 for newest improvements.
A faster, lower-cost GPT-5 for well-defined tasks. Text & vision with long context at a fraction of the price.
Fastest, most cost-efficient GPT-5 variant.
Open-weight model entry as listed by OpenAI (see models page). Token costs depend on where you run it.
Open-weight model entry as listed by OpenAI (see models page). Token costs depend on where you run it.
Alias tier aligned with GPT-5 Mini pricing; use when your workflow targets the "low" cost tier.
Mid-tier GPT-5 variant balancing performance and cost for general-purpose workloads.
High-tier GPT-5 with enhanced reasoning, reliability and coding for mission-critical workloads.
A 1T parameter open-weight Mixture-of-Experts (MoE) model with 32B active parameters. This is the unaligned, pre-trained base model, suitable for further fine-tuning.
The instruction-tuned version of Kimi K2, optimized for chat, agentic tasks, and tool use. Aligned with RLHF for helpful and safe responses.
A 30B parameter model from the Qwen3 series, excelling in coding and agentic tasks with a 1M token context length.
GLM-4.5 agentic foundation model. Official pricing is published in RMB on BigModel; USD prices vary by provider.
Qwen3 coding-specialized model with long-context capabilities and strong tool-use.
Version of o3 with more compute for better responses (Responses API only).
Anthropic's most powerful model, excelling in coding, advanced reasoning, and AI agent workflows. Handles complex, long-running tasks.
Anthropic's highly capable and versatile model, offering a strong balance of intelligence, speed, and cost-effectiveness for enterprise applications.
Google's most advanced reasoning Gemini model, capable of solving complex problems. Supports text, code, image, audio, and video inputs. Features a 1M token context window (up to 2M in some versions).
Mistral AI's frontier-class multimodal model balancing SOTA performance, lower cost, and simpler deployability for enterprise usage. Excels in coding and multimodal understanding.
A 24B open-source text model from Mistral AI that excels at using tools to explore codebases, editing multiple files, and powering software engineering agents. Apache 2.0 license.
Google's best model for price and performance (as of May 2025), featuring hybrid reasoning capabilities. Supports text, code, image, audio, and video inputs. 1M token context window.
OpenAI's fast coding model designed for the Codex coding agent. Optimized for rapid code generation, editing, and review within development workflows.
Fast and creative video generation with emphasis on artistic styles and special effects. Free tier available with paid options.
OpenAI reasoning model for complex tasks (text+image input, text output). Succeeded by GPT-5.x for many agentic workloads.
A faster, cost-efficient reasoning model, successor to o3-mini, released in April 2025. Offers strong performance on math, coding, and vision. Can process text and images, and features autonomous tool use.
Alibaba's flagship Qwen3 Mixture-of-Experts model with 235B total parameters (22B active). Features hybrid reasoning and supports 119 languages. (Note: Not publicly available at release).
Alibaba's Qwen3 Mixture-of-Experts model with 30B total parameters (3B active). Features hybrid reasoning and supports 119 languages. Apache 2.0 license.
Alibaba's largest dense model in the Qwen3 family with 32B parameters. Features hybrid reasoning and supports 119 languages. Apache 2.0 license.
Smartest non-reasoning GPT model (text+image in, text out).
Smaller, faster GPT-4.1 tier with low cost.
Fastest, cheapest GPT-4.1 tier.
Preview of Google's Gemini 2.5 Flash with hybrid reasoning capabilities. Ultra-fast and cost-efficient for high-volume applications.
Meta's Llama 4 Maverick with mixture-of-experts architecture, 1M context window, and strong multilingual support. Open-source model.
Meta's Llama 4 Scout with an industry-leading 10M token context window and 16 experts MoE architecture. Optimized for efficiency.
A new leader in the small models category by Mistral AI, with image understanding capabilities and an extended 128k context length. Apache 2.0 license.
GPT-4o variant with built-in web search grounding. Provides up-to-date, cited answers by searching the web in real time.
Early preview of Google's Gemini 2.5 Pro thinking model. Excels at reasoning, coding, and multimodal tasks with a 1M token context window.
Cohere's next-generation enterprise model with an expanded 256K context window. Optimized for agentic RAG, tool use, and structured outputs.
Google's open-source 27B parameter model from the Gemma 3 family. Natively multimodal with strong performance on text, image, and video tasks. Free to use under open license.
Hybrid reasoning Claude 3.x model (extended thinking). Deprecated; recommended replacement is Claude Sonnet 4.5.
A faster, more cost-effective version of o3, released in January 2025. Offers strong reasoning, coding, and vision capabilities. Optimized for math and coding tasks.
A powerful and efficient model from Mistral AI for languages from the Middle East and South Asia.
xAI's flagship large language model with strong reasoning, coding, and math capabilities. Trained on the Colossus supercluster.
xAI's lightweight reasoning model with think mode. Faster and more cost-efficient than Grok 3 while maintaining strong reasoning capabilities.
DeepSeek's enhanced model with improved reasoning capabilities, expanded context window, and even more competitive pricing.
Mistral AI's cutting-edge language model for coding (second version). Specializes in low-latency, high-frequency tasks like fill-in-the-middle (FIM), code correction, and test generation.
DeepSeek's reasoning model trained with reinforcement learning. Excels at math, coding, and complex reasoning tasks with transparent chain-of-thought.
Google's latest experimental model with breakthrough multimodal capabilities and enhanced reasoning at an extremely competitive price point.
Meta's latest 70B parameter model with improved performance and capabilities, offering state-of-the-art results for its size.
DeepSeek's latest model with strong performance across reasoning, coding, and general tasks at competitive pricing.
Previous full o-series reasoning model (text+image in, text out).
Higher-compute variant of o1 for better responses.
Microsoft's small language model with 14B parameters that punches well above its weight. Excels at STEM reasoning and coding despite its compact size. Open source under MIT license.
Amazon's highly capable multimodal model balancing accuracy, speed, and cost. Processes text, images, and video inputs for a wide range of enterprise tasks.
Amazon's very low-cost multimodal model for high-volume tasks. Processes text, images, and video at extremely competitive pricing via Amazon Bedrock.
Advanced text-to-video generation with realistic motion and scene understanding. Pricing per video generation varies by length and resolution.
Anthropic's most advanced model, significantly improved over Claude 3 Sonnet with enhanced reasoning, coding, and vision capabilities.
Claude 3.5 Haiku snapshot. Deprecated; recommended replacement is Haiku 4.5.
Deprecated preview snapshot of OpenAI's first o-series reasoning model. Kept for backwards compatibility; prefer o1 for production.
Smaller, faster o-series reasoning model. Deprecated in favor of newer reasoning models, but still supported in legacy workflows.
Meta's multimodal model combining text and vision capabilities with strong performance across various tasks.
A smaller, efficient multimodal model from Meta with vision capabilities, suitable for edge deployment and cost-sensitive applications.
Mistral AI's cost-effective model for straightforward tasks, offering good performance and efficiency.
Alibaba's large language model with strong performance in reasoning, coding, and multilingual tasks.
AI21's large hybrid SSM-Transformer model with a 256K context window. Uses a novel Jamba architecture combining Mamba SSM layers with Transformer attention for efficient long-context processing.
OpenAI's most affordable and fastest model in the GPT-4o family, designed for high-volume, low-latency tasks.
Meta's largest and most capable Llama 3.1 model, designed for complex reasoning, coding, and nuanced instruction following.
A large instruction-tuned model from Meta's Llama 3.1 series, offering a strong balance of performance and efficiency for a wide range of tasks.
A highly efficient instruction-tuned model from Meta's Llama 3.1 series, suitable for fast, on-device, or edge applications.
Mistral AI's flagship model with enhanced reasoning, coding, and multilingual capabilities.
Professional video generation and editing AI with motion control and style consistency. Subscription-based pricing.
OpenAI's flagship multimodal model, natively processing text, audio, and images for faster, more capable interactions.
Google's faster and lower-cost version of Gemini 1.5 Pro, optimized for high-volume, high-frequency tasks while retaining a large context window and multimodal capabilities.
Mistral AI's open-weight generative model specialized for code generation, supporting 80+ languages.
OpenAI's powerful model prior to GPT-4o, with a large context window and strong performance on complex tasks. Supports vision.
Cohere's most powerful model optimized for enterprise RAG and tool use. Excels at grounded generation with citations and multi-step tool workflows.
Claude 3 Opus (deprecated) - previous highest-intelligence Claude 3 model.
A balanced model from Anthropic, offering a blend of intelligence and speed, ideal for enterprise workloads and scaled AI deployments.
Anthropic's fastest and most compact model, designed for near-instant responsiveness and high throughput tasks.
Google's highly capable multimodal model with a breakthrough long context window of up to 2 million tokens. Excels at complex reasoning, problem-solving, and understanding long-form content.
Game asset and creative image generation with consistent character and style creation. Credit-based pricing system.
State-of-the-art text-to-image generation with improved prompt following and image quality. Pricing per image: HD 1024×1024 $0.040, Standard $0.020
Open-source image generation model with fine-tuning capabilities. API pricing varies by provider, self-hosting available.
OpenAI's fast, cost-effective model optimized for chat and simple tasks.
Anthropic's flagship Opus-class model for July 2026. Opus 5 pairs Mythos-class reasoning with the Opus price point, adds adaptive effort control and a hardened long-horizon agent loop that holds context across day-long tasks.
The 1M-token context tier of Claude Opus 5. Same model, long-context premium pricing applied to requests above 200K input tokens. Built for whole-repository and whole-corpus work.
The workhorse of the Claude 5 family. Sonnet 5 lands close to Opus 4.8 on coding and agentic benchmarks while staying at the long-standing Sonnet price, with a 1M token context window available by default.
OpenAI's Sol tier for GPT-5.6: a long-thinking configuration that spends far more reasoning tokens per request and tops OpenAI's own agentic and math evaluations. Priced as a premium reasoning endpoint.
The July 2026 refresh of the GPT-5 series. GPT-5.6 keeps GPT-5.5 pricing while improving instruction following, tool-call accuracy and refusal calibration, and it ships with a faster streaming stack.
The cost tier of GPT-5.6. Mini keeps the same tokenizer and tool interface as the full model, trading depth for roughly an eighth of the price and noticeably lower latency.
The Codex-tuned build of GPT-5.6 for the Codex CLI, IDE extension and cloud agents. Optimised for long autonomous coding sessions, patch generation and test repair rather than open-ended chat.
Google's flagship Gemini for mid-2026. Gemini 3.6 Pro extends the 2M token window to all paid tiers, improves grounded answers with live Search, and adds native 1 fps video understanding.
The fast, cheap tier of Gemini 3.6. Flash keeps a 1M token window and multimodal input while running several times faster than Pro, which makes it the default choice for high-volume production traffic.
Moonshot's flagship Kimi K3.7, an open-weight mixture-of-experts model with a 512K token window. K3.7 closes most of the gap to Western frontier models on agentic coding while staying dramatically cheaper.
The low-latency hosted tier of Kimi K3.7. Same weights, dedicated high-throughput serving, roughly double the price and a large jump in tokens per second.
A distilled K3.7 aimed at classification, extraction and sub-agent duty. Runs comfortably on a single high-memory GPU when self-hosted.
Alibaba's Qwen 3.8 flagship. A sparse MoE model with 256K context, first-class support for more than 100 languages and a notable jump in tool-calling reliability over Qwen 3.
The coding-specialised Qwen 3.8 with a 1M token window. Tuned for repository-scale understanding, patch generation and terminal agents.
Anthropic's most powerful frontier model - the Mythos-class system with state-of-the-art agentic and cybersecurity capability. Distributed under strict access controls; Fable 5 is its safeguarded public sibling.
The first public Mythos-class model - Anthropic's most powerful generally available model, sharing Mythos 5's underlying intelligence with added safeguards that fall back to Opus 4.8 on sensitive prompts.
Meta's flagship open-weight model for 2026. Llama 5 Maverick is a large mixture-of-experts system with a 2M token window and native multimodal input, released under the Llama 5 Community License.
The compact member of the Llama 5 family. Scout runs on a single 24 GB consumer GPU at full speed while keeping a 1M token window and vision input, which makes it the practical choice for local and edge work.
Meta's largest released model, positioned against closed frontier systems. Behemoth is served through partners rather than being practical to self-host, and it leads open-weight benchmarks on reasoning and math.
A mid-cycle refresh of DeepSeek V4 with a doubled context window, better tool calling and a large improvement in multilingual output quality. Still open weights.
xAI's Grok 4 with a 1M token window, live X and web grounding, and a large jump in coding benchmarks over Grok 3.5. Available through the xAI API and X Premium+.
The cost tier of Grok 4. Keeps live grounding and the 1M window but targets high-volume traffic at roughly a tenth of the flagship price.
Anthropic's most capable Opus-tier model - highly autonomous, state-of-the-art on long-horizon agentic work, knowledge work and memory, with clearer, warmer writing. 1M context at standard pricing.
Google's newest fast model (Google I/O 2026), built for agentic tasks and coding - beats Gemini 3.1 Pro on coding and agentic benchmarks at a fraction of the cost.
Mistral's refreshed flagship with a 256K window, stricter JSON mode and EU data residency through La Plateforme. The default pick for European teams with GDPR locality requirements.
Zhipu's GLM-5, an open-weight MoE model with strong bilingual Chinese and English performance and a competitive agentic coding score at a fraction of frontier pricing.
OpenAI's flagship GPT-5.5, released April 2026. A ~1.05M token context window with strong terminal/agentic benchmarks; available in the Responses and Chat Completions APIs.
A higher-accuracy GPT-5.5 variant for the most demanding tasks, available via the API at premium pricing.
OpenAI's most advanced model with enhanced reasoning, longer context, and multimodal capabilities. Top of the line for complex tasks.
Affordable version of GPT-5.4 with strong performance for everyday tasks at a fraction of the cost.
Fast and affordable Gemini 3.1 Flash optimized for high-throughput applications at minimal cost.
xAI's latest flagship model with expanded context, improved reasoning, and deeper integration with real-time data from X platform.
Previous-generation Opus - highly autonomous with excellent long-horizon agentic work, vision and memory. Adaptive thinking only; high-resolution vision support.
DeepSeek's second-generation reasoning model. R2 exposes its full chain of thought, scores near the top of competitive math benchmarks and is priced far below comparable Western reasoning endpoints.
Anthropic's Claude Opus 4.6 with extended 1M token context window for processing entire codebases, books, and massive datasets in a single prompt.
Anthropic's latest flagship model with best-in-class reasoning, coding, and agentic capabilities. Supports extended thinking for complex multi-step problems.
Anthropic's balanced mid-tier model offering strong intelligence, speed, and cost-effectiveness. Excellent for everyday coding and enterprise tasks.
Google's most capable Gemini Pro model with an industry-leading 2M token context window - ideal for large-document analysis and long multi-turn reasoning.
DeepSeek's fourth-generation open-weights model with state-of-the-art reasoning at remarkably low cost. Strong performance on coding and math benchmarks.
An o3-family model optimized for deep research tasks. Autonomously browses the web, synthesizes information, and produces comprehensive research reports.
Mistral's latest flagship model with multilingual excellence, strong coding, and enterprise-grade function calling. Open-weight model.
Alibaba's latest Qwen 3 flagship model with strong multilingual capabilities and improved reasoning. Competitive with frontier models at lower cost.
Professional AI image generation with photorealistic quality and artistic control. Subscription-based: Basic $10/mo, Standard $30/mo, Pro $60/mo
Higher-compute variant of GPT-5.2 for harder problems (Responses API only).
Most intelligent Codex model optimized for long-horizon agentic coding (Responses API only).
Gemini 3 Flash Preview on Vertex AI.
OpenAI flagship model for coding and agentic tasks across industries.
Flagship GPT model with configurable reasoning effort; predecessor to GPT-5.2.
GPT-5.1 variant optimized for agentic coding in Codex (Responses API only).
Claude 4.5 flagship Opus model for long-horizon coding and agentic workflows.
Gemini 3 Pro Preview on Vertex AI.
Anthropic's fastest 4.5-series model-cost-effective for high-volume workloads with extended thinking and strong tool use.
Windsurf/Cognition in-house frontier model for agentic coding. Consumed via Windsurf prompt credits (not USD per-token).
Anthropic's fastest and most cost-effective 4.5-series model, optimized for high-volume workloads with strong tool use capabilities.
A version of GPT-5 optimized for agentic coding in Codex. Default in Codex cloud & reviews; also usable via API key. Priced the same as GPT-5.
Anthropic's most intelligent model for agents and coding, with extended thinking and state-of-the-art performance on SWE-bench Verified.
Next-generation Gemini with advanced multimodal reasoning and long context. Rates pending official pricing page.
Opus 4.1 (200k) - higher-cost flagship Opus tier. Newer Opus 4.5 provides a more accessible price point.
Version of GPT-5 that produces smarter and more precise responses. Responses API only; higher max output than standard GPT-5.
Previous flagship GPT model for coding, reasoning, and agentic tasks. OpenAI recommends GPT-5.1/5.2 for newest improvements.
A faster, lower-cost GPT-5 for well-defined tasks. Text & vision with long context at a fraction of the price.
Fastest, most cost-efficient GPT-5 variant.
Open-weight model entry as listed by OpenAI (see models page). Token costs depend on where you run it.
Open-weight model entry as listed by OpenAI (see models page). Token costs depend on where you run it.
Alias tier aligned with GPT-5 Mini pricing; use when your workflow targets the "low" cost tier.
Mid-tier GPT-5 variant balancing performance and cost for general-purpose workloads.
High-tier GPT-5 with enhanced reasoning, reliability and coding for mission-critical workloads.
A 1T parameter open-weight Mixture-of-Experts (MoE) model with 32B active parameters. This is the unaligned, pre-trained base model, suitable for further fine-tuning.
The instruction-tuned version of Kimi K2, optimized for chat, agentic tasks, and tool use. Aligned with RLHF for helpful and safe responses.
A 30B parameter model from the Qwen3 series, excelling in coding and agentic tasks with a 1M token context length.
GLM-4.5 agentic foundation model. Official pricing is published in RMB on BigModel; USD prices vary by provider.
Qwen3 coding-specialized model with long-context capabilities and strong tool-use.
Version of o3 with more compute for better responses (Responses API only).
Anthropic's most powerful model, excelling in coding, advanced reasoning, and AI agent workflows. Handles complex, long-running tasks.
Anthropic's highly capable and versatile model, offering a strong balance of intelligence, speed, and cost-effectiveness for enterprise applications.
Google's most advanced reasoning Gemini model, capable of solving complex problems. Supports text, code, image, audio, and video inputs. Features a 1M token context window (up to 2M in some versions).
Mistral AI's frontier-class multimodal model balancing SOTA performance, lower cost, and simpler deployability for enterprise usage. Excels in coding and multimodal understanding.
A 24B open-source text model from Mistral AI that excels at using tools to explore codebases, editing multiple files, and powering software engineering agents. Apache 2.0 license.
Google's best model for price and performance (as of May 2025), featuring hybrid reasoning capabilities. Supports text, code, image, audio, and video inputs. 1M token context window.
OpenAI's fast coding model designed for the Codex coding agent. Optimized for rapid code generation, editing, and review within development workflows.
Fast and creative video generation with emphasis on artistic styles and special effects. Free tier available with paid options.
OpenAI reasoning model for complex tasks (text+image input, text output). Succeeded by GPT-5.x for many agentic workloads.
A faster, cost-efficient reasoning model, successor to o3-mini, released in April 2025. Offers strong performance on math, coding, and vision. Can process text and images, and features autonomous tool use.
Alibaba's flagship Qwen3 Mixture-of-Experts model with 235B total parameters (22B active). Features hybrid reasoning and supports 119 languages. (Note: Not publicly available at release).
Alibaba's Qwen3 Mixture-of-Experts model with 30B total parameters (3B active). Features hybrid reasoning and supports 119 languages. Apache 2.0 license.
Alibaba's largest dense model in the Qwen3 family with 32B parameters. Features hybrid reasoning and supports 119 languages. Apache 2.0 license.
Smartest non-reasoning GPT model (text+image in, text out).
Smaller, faster GPT-4.1 tier with low cost.
Fastest, cheapest GPT-4.1 tier.
Preview of Google's Gemini 2.5 Flash with hybrid reasoning capabilities. Ultra-fast and cost-efficient for high-volume applications.
Meta's Llama 4 Maverick with mixture-of-experts architecture, 1M context window, and strong multilingual support. Open-source model.
Meta's Llama 4 Scout with an industry-leading 10M token context window and 16 experts MoE architecture. Optimized for efficiency.
A new leader in the small models category by Mistral AI, with image understanding capabilities and an extended 128k context length. Apache 2.0 license.
GPT-4o variant with built-in web search grounding. Provides up-to-date, cited answers by searching the web in real time.
Early preview of Google's Gemini 2.5 Pro thinking model. Excels at reasoning, coding, and multimodal tasks with a 1M token context window.
Cohere's next-generation enterprise model with an expanded 256K context window. Optimized for agentic RAG, tool use, and structured outputs.
Google's open-source 27B parameter model from the Gemma 3 family. Natively multimodal with strong performance on text, image, and video tasks. Free to use under open license.
Hybrid reasoning Claude 3.x model (extended thinking). Deprecated; recommended replacement is Claude Sonnet 4.5.
A faster, more cost-effective version of o3, released in January 2025. Offers strong reasoning, coding, and vision capabilities. Optimized for math and coding tasks.
A powerful and efficient model from Mistral AI for languages from the Middle East and South Asia.
xAI's flagship large language model with strong reasoning, coding, and math capabilities. Trained on the Colossus supercluster.
xAI's lightweight reasoning model with think mode. Faster and more cost-efficient than Grok 3 while maintaining strong reasoning capabilities.
DeepSeek's enhanced model with improved reasoning capabilities, expanded context window, and even more competitive pricing.
Mistral AI's cutting-edge language model for coding (second version). Specializes in low-latency, high-frequency tasks like fill-in-the-middle (FIM), code correction, and test generation.
DeepSeek's reasoning model trained with reinforcement learning. Excels at math, coding, and complex reasoning tasks with transparent chain-of-thought.
Google's latest experimental model with breakthrough multimodal capabilities and enhanced reasoning at an extremely competitive price point.
Meta's latest 70B parameter model with improved performance and capabilities, offering state-of-the-art results for its size.
DeepSeek's latest model with strong performance across reasoning, coding, and general tasks at competitive pricing.
Previous full o-series reasoning model (text+image in, text out).
Higher-compute variant of o1 for better responses.
Microsoft's small language model with 14B parameters that punches well above its weight. Excels at STEM reasoning and coding despite its compact size. Open source under MIT license.
Amazon's highly capable multimodal model balancing accuracy, speed, and cost. Processes text, images, and video inputs for a wide range of enterprise tasks.
Amazon's very low-cost multimodal model for high-volume tasks. Processes text, images, and video at extremely competitive pricing via Amazon Bedrock.
Advanced text-to-video generation with realistic motion and scene understanding. Pricing per video generation varies by length and resolution.
Anthropic's most advanced model, significantly improved over Claude 3 Sonnet with enhanced reasoning, coding, and vision capabilities.
Claude 3.5 Haiku snapshot. Deprecated; recommended replacement is Haiku 4.5.
Deprecated preview snapshot of OpenAI's first o-series reasoning model. Kept for backwards compatibility; prefer o1 for production.
Smaller, faster o-series reasoning model. Deprecated in favor of newer reasoning models, but still supported in legacy workflows.
Meta's multimodal model combining text and vision capabilities with strong performance across various tasks.
A smaller, efficient multimodal model from Meta with vision capabilities, suitable for edge deployment and cost-sensitive applications.
Mistral AI's cost-effective model for straightforward tasks, offering good performance and efficiency.
Alibaba's large language model with strong performance in reasoning, coding, and multilingual tasks.
AI21's large hybrid SSM-Transformer model with a 256K context window. Uses a novel Jamba architecture combining Mamba SSM layers with Transformer attention for efficient long-context processing.
OpenAI's most affordable and fastest model in the GPT-4o family, designed for high-volume, low-latency tasks.
Meta's largest and most capable Llama 3.1 model, designed for complex reasoning, coding, and nuanced instruction following.
A large instruction-tuned model from Meta's Llama 3.1 series, offering a strong balance of performance and efficiency for a wide range of tasks.
A highly efficient instruction-tuned model from Meta's Llama 3.1 series, suitable for fast, on-device, or edge applications.
Mistral AI's flagship model with enhanced reasoning, coding, and multilingual capabilities.
Professional video generation and editing AI with motion control and style consistency. Subscription-based pricing.
OpenAI's flagship multimodal model, natively processing text, audio, and images for faster, more capable interactions.
Google's faster and lower-cost version of Gemini 1.5 Pro, optimized for high-volume, high-frequency tasks while retaining a large context window and multimodal capabilities.
Mistral AI's open-weight generative model specialized for code generation, supporting 80+ languages.
OpenAI's powerful model prior to GPT-4o, with a large context window and strong performance on complex tasks. Supports vision.
Cohere's most powerful model optimized for enterprise RAG and tool use. Excels at grounded generation with citations and multi-step tool workflows.
Claude 3 Opus (deprecated) - previous highest-intelligence Claude 3 model.
A balanced model from Anthropic, offering a blend of intelligence and speed, ideal for enterprise workloads and scaled AI deployments.
Anthropic's fastest and most compact model, designed for near-instant responsiveness and high throughput tasks.
Google's highly capable multimodal model with a breakthrough long context window of up to 2 million tokens. Excels at complex reasoning, problem-solving, and understanding long-form content.
Game asset and creative image generation with consistent character and style creation. Credit-based pricing system.
State-of-the-art text-to-image generation with improved prompt following and image quality. Pricing per image: HD 1024×1024 $0.040, Standard $0.020
Open-source image generation model with fine-tuning capabilities. API pricing varies by provider, self-hosting available.
OpenAI's fast, cost-effective model optimized for chat and simple tasks.