OpenAI's flagship multimodal model that can reason across text, images, and audio. Offers strong performance across a wide range of tasks with fast response times.

The latest evolution of the GPT-4 family with improved instruction following, coding, and long-context performance. Supports up to 1M token context window.

OpenAI's advanced reasoning model using chain-of-thought to solve complex problems in math, science, and coding.

RETIRED May 12, 2026 — the API now returns errors. Kept for historical comparison; the GPT Image line (GPT Image 2.5) is the official replacement. Was known for high-quality images with strong prompt following and ChatGPT integration.

Open-source speech recognition model trained on 680,000 hours of multilingual audio. Capable of transcription and translation across 99 languages. The hosted whisper-1 API was deprecated Aug 26, 2026 (shutdown Feb 26, 2027; replacement: gpt-transcribe) — self-hosted open weights remain available.

Anthropic's most powerful frontier model to date, internally codenamed 'Capybara'. Uses Mixture of Experts architecture with an estimated ~10 trillion parameters. Restricted to vetted organizations through Project Glasswing due to extraordinary cybersecurity capabilities deemed too dangerous for public release.

Anthropic's most capable model, excelling at complex analysis, nuanced writing, coding, and multi-step reasoning. Known for strong safety alignment.

Anthropic's balanced model offering strong performance at lower cost than Opus. Ideal for most production workloads.

Anthropic's fastest and most affordable model. Designed for high-throughput, low-latency tasks like classification and extraction.

Google's most capable model with a massive 1M token context window. Excels at long-document analysis and multimodal tasks.

Google's fast, efficient model optimized for speed and cost while maintaining strong quality.

Google's open-weight model family derived from Gemini research. Available in sizes from 1B to 27B for on-device and self-hosted deployments. Superseded by Gemma 4 (March 2026) but still widely used and supported.

Meta's latest open-weight Mixture-of-Experts model with 400B total parameters (17B active). Excellent multilingual performance.

Meta's high-performance dense model offering near-405B quality at a fraction of the cost. Popular for self-hosting and fine-tuning.

Mistral's flagship model with 128K context window. Strong at multilingual tasks, code generation, and reasoning. Superseded by Mistral Large 3 (Dec 2025) — open-weight Apache 2.0, multimodal, at $0.50/$1.50 per MTok.

Mistral's cost-efficient open-weight model optimized for simple tasks and high-throughput applications. Apache 2.0 license. Superseded by Mistral Small 4 (v26.03) — hybrid instruct/reasoning/coding, Apache 2.0, $0.15/$0.60 per MTok.

Mistral's specialized code generation model optimized for code completion and understanding across 80+ programming languages.

Powerful open-source MoE model with 671B total parameters, known for exceptional cost-efficiency. No longer served by the official DeepSeek API — replaced by the V4 series (currently DeepSeek-V4.1-Flash); open weights remain available for self-hosting.

DeepSeek's reasoning model rivaling OpenAI o1. Uses chain-of-thought for complex math, science, and coding. Fully open source. No longer served by the official DeepSeek API (the deepseek-reasoner alias was retired in July 2026) — reasoning now lives in the V4/V4.1 thinking mode; open weights remain available.

xAI's former flagship with strong reasoning and real-time information access via X (Twitter). No longer listed on the official xAI API models page — superseded by the Grok 4.x line (current flagship: Grok 4.6).

Cohere's enterprise-focused model optimized for RAG, tool use, and business applications. Strong at grounded generation with citations. Superseded by Command A (March 2025, 111B, 256K context) and the open-weight flagship Command A+ (May 2026, Apache 2.0).

The most popular open-source image generation model with extensive customization through LoRAs, ControlNet, and community models.

Leading AI image generation service known for stunning artistic quality. Widely used by artists, designers, and creative professionals. The v6 entry is two generations behind: V8.1 became the default in June 2026 and V8.2 (July 24, 2026) is the current default model.

State-of-the-art open-source image generation model from the creators of Stable Diffusion. Excellent prompt following and quality. The FLUX.1 line is still sold, but BFL has since shipped the FLUX.2 family (January 2026, including the open-weight 32B FLUX.2 [dev]) and FLUX 3 (July 2026), a multimodal image+video+audio foundation model.

OpenAI's video generation model, now being shut down: the Sora app closed April 26, 2026 and the sora-2 / sora-2-pro API shuts down September 24, 2026 with no OpenAI replacement. Kept for historical comparison.

OpenAI's most capable embedding model for converting text into vectors. Used for semantic search, clustering, and RAG.

Alibaba's flagship open-source model with strong multilingual performance, excelling in Chinese and English.

Microsoft's compact but powerful model designed to maximize quality-per-parameter. Excels at reasoning and coding despite its small size.

OpenAI's small, affordable, and fast model. Outperforms GPT-3.5 Turbo at a fraction of GPT-4o's cost. Ideal for lightweight tasks.

Smaller, faster, cheaper variant of GPT-4.1 balancing cost and performance. Strong instruction following at a fraction of the cost.

The smallest and fastest GPT-4.1 variant. Optimized for speed and cost, suitable for simple tasks at massive scale.

OpenAI's cost-efficient reasoning model. Brings chain-of-thought reasoning at a fraction of o3's cost, ideal for applications that need reasoning without the premium price.

Smaller, cheaper version of OpenAI's o3 reasoning model. Good balance of reasoning ability and cost for everyday tasks requiring logical thinking.

Anthropic's fastest and most affordable current model, released October 2025. Compact yet capable with extended thinking support, ideal for lightweight tasks, real-time responses, subagents, and cost-sensitive deployments.

RETIRED — shut down June 1, 2026 (including Flash-Lite). Kept for historical comparison. Was a fast, efficient model for high-throughput applications; migrate to Gemini 2.5 Flash or the Gemini 3.x Flash line.

Google's open-source lightweight model family. Efficient and capable, designed for on-device and resource-constrained deployments.

Meta's efficient open-source model using mixture-of-experts architecture. 17B active parameters from 109B total, offering strong performance at low compute cost.

xAI's compact reasoning model with chain-of-thought capabilities. Fast and affordable alternative to full Grok 3 with built-in reasoning. No longer listed on the official xAI API models page — superseded by the Grok 4.x line.

Leading AI voice synthesis platform offering realistic text-to-speech, voice cloning, and dubbing. The current flagship is Eleven v3 (70+ languages), joined in 2026 by Eleven v3 Conversational (~280ms realtime speech for agents, August 2026), Scribe v2 speech-to-text, and Eleven Music v2.5 (September 2026).

AI search model that combines LLM capabilities with real-time web search, returning cited, up-to-date answers. DEPRECATED: the Sonar Chat Completions API is supported only until September 27, 2026 — Perplexity directs developers to the new Agent API. Request fees are now tiered at $5-12 per 1K requests by search context size.

Image generation model excelling at design-quality output, formerly a top performer on ELO rankings. Superseded by Recraft V4 (February 2026) and the current flagship V4.1 (May 2026, $0.035/image, Pro $0.21); V3 remains available via API at $0.04/image.

Mistral's balanced model offering strong performance between Small and Large. Good all-rounder for diverse enterprise tasks with European data residency. Superseded by Mistral Medium 3.5 (May 2026) — frontier multimodal with agentic/coding focus at $1.50/$7.50 per MTok.

High-performance multilingual embedding model supporting 89 languages, optimized for search, retrieval, and RAG. Two generations behind: Jina has since shipped jina-embeddings-v4 (June 2025), v5-text (February 2026), and v5-omni (May 2026, the first text/image/audio/video/PDF embeddings); v3 is still offered but no longer recommended.

LEGACY — the moonshot-v1 series and Kimi K2.5 were retired from the API on August 31, 2026 (requests now return 404); migrate to Kimi K3. Was known for one of the longest context windows available and strong long-document understanding.

Video generation model from Kuaishou with 1080p HDR output. Produces high-quality videos from text or images with superior motion, physics, and facial expression understanding. Superseded by the Kling 3.0 family (2026), including Kling 3.0 Turbo (June 2026) with multi-shot clips and native audio with lip-sync.

RETIRED — shut down November 10, 2025; its successor Imagen 4 was also shut down August 17, 2026. The Imagen line is replaced by Gemini-native image models (Nano Banana). Kept for historical comparison.

Amazon's multimodal model offering strong performance across text, image, and video understanding, optimized for accuracy, speed, and cost on AWS. Superseded by the Nova 2 generation (December 2025): Nova 2 Lite is GA and Nova 2 Pro (1M context) remains in preview as of September 2026.

01.AI's fast and capable model rivaling GPT-4o class performance. Strong on benchmarks with competitive pricing and fast inference.

Alibaba's open-source model family with dense and MoE variants. Strong multilingual performance rivaling frontier closed-source models. Superseded by the Qwen 3.8 generation (August 2026), including the open-weight Qwen3.8-27B.

OpenAI frontier model for complex professional work with a 400K token context window and 128K max output. Strong reasoning, coding, and multimodal capabilities.

Anthropic model that handles ambiguity and complex multi-system bugs without hand-holding. Excels at long-horizon autonomous tasks with sustained reasoning and fewer dead-ends.

Google's reasoning-focused model with internal chain-of-thought. Breaks problems into steps, checks its own logic, and refines output. Supports text, vision, and native image generation. Superseded by Gemini 3.1 Pro Preview (Feb 2026); Gemini 3.5 Pro was announced at I/O May 2026 but remains unreleased.

Google DeepMind image generation model built on Gemini 3, generally available since May 28, 2026 as `gemini-3-pro-image` (Nano Banana Pro). Offers personalized image generation using Google Photos context and high-fidelity text rendering for professional asset production. API pricing: $2/1M input, $120/1M image output (~$0.134 per 1K/2K image, $0.24 per 4K).

OpenAI's natively multimodal image generation model. Accepts both text and image inputs through a unified transformer, enabling seamless text-to-image and image editing.

Next-generation 32B parameter image generation model with up to 4 megapixel output. Features the Kontext Engine for natural language image editing and dramatically improved text rendering. BFL has since announced FLUX 3 (July 2026), a multimodal foundation model whose Video API (August 2026) generates up to 20s clips with synchronized audio from $0.17/s.

xAI's model with video and image generation capabilities via the Imagine API. Supports text-to-video, image-to-video, and synchronized audio generation at 720p. As an LLM, Grok 4 is superseded by Grok 4.5 and Grok 4.6 (2026); the Imagine image/video models remain on the official API.

Google Gemini Flash optimized for audio processing tasks. Transcribes, analyzes, and reasons about audio content with fast response times and competitive pricing.

Google's real-time multimodal streaming API for voice-in, voice-out conversational AI. Supports low-latency audio interaction with function calling mid-conversation.

OpenAI's text-to-speech API with multiple quality tiers and 13+ voices. Converts text to natural-sounding speech with real-time streaming support.

Emotionally intelligent voice AI platform with the Empathic Voice Interface (EVI). Offers expressive TTS with emotion detection, natural language tone control, and under 200ms latency.

Lightweight open-source TTS model with 82M parameters delivering quality comparable to commercial models. Supports 9 languages, 54+ voices, and runs on both GPU and CPU.

Image generation model that topped the Artificial Analysis leaderboard. Produces up to 4K resolution images with precise text rendering, custom fonts, and self-correction pipeline. Superseded by Riverflow 2.5 Pro and 2.5 Fast (June 2026), which add agentic multi-step editing with an internal judge and up to 10 reference images.

ByteDance's fast and cost-effective image generation model. Generates images at ~1.8 seconds with strong text rendering accuracy and multi-reference support. Superseded by Seedream 5.0 Pro (July 2026, ~$0.045/image) with layer separation and interactive editing.

Google DeepMind's video generation model family, currently at Veo 3.1 (preview). Generates high-quality videos from text prompts with audio, in tiers from Lite ($0.05-0.08/s) through Fast ($0.10-0.30/s) to Standard ($0.40/s, $0.60/s in 4K). Veo 2.0 and 3.0 were shut down June 30, 2026.

OpenAI's most capable frontier model optimized for agentic, coding, and professional workflows. Features native computer-use capabilities, 1M token context, and the most token-efficient reasoning of the GPT-5 family.

OpenAI's most capable agentic coding model combining frontier coding performance with deep reasoning. Handles long-running tasks involving research, tool use, and complex execution. Can be steered interactively while working.

OpenAI's efficient and affordable embedding model for converting text into 1536-dimensional vectors. Ideal for semantic search, clustering, and RAG with a 5x price reduction over ada-002.

Google's high-efficiency image generation and editing model (Nano Banana 2), generally available since May 28, 2026 as `gemini-3.1-flash-image` — the preview endpoints were shut down August 17, 2026. Delivers Pro-level visual quality at Flash speed: text-to-image, image-to-image editing, multi-turn conversational editing, and up to 4K output.

Anthropic's frontier model for the most demanding reasoning and long-horizon agentic work, with exceptional performance in software engineering, scientific research, and vision tasks. Still available, but superseded by Claude Fable 5.1 (September 2026) at the same price.

Anthropic's recommended model for complex agentic coding and enterprise work. Offers a 1M context window and adaptive thinking that defaults to high effort, making it the current flagship for demanding multi-step tasks.

Anthropic's best combination of speed and intelligence. 1M token context with adaptive thinking and fast response times, at $2 per million input tokens (the introductory price was made permanent in August 2026).

Anthropic's flagship Opus-tier model, released July 2026. Within 0.5% of Claude Fable 5 on agentic coding benchmarks at half the cost, with adaptive thinking on by default and state-of-the-art results in debugging, computer use, and long self-verifying tasks.

Anthropic's most capable widely available model, released September 2026 as the successor to Claude Fable 5. Built for long-running agentic coding, multistep research, and 1M-context work, with cache reads 75% cheaper than Fable 5 and built-in text watermarking with C2PA content credentials.

Former OpenAI flagship designed for long-horizon agentic tasks, excelling at coding, research, data analysis, and computer use. Superseded as flagship by the GPT-5.6 family (July 2026) and GPT-6 Astra (September 2026), but still available.

Flagship tier of the GPT-5.6 family (Sol/Terra/Luna), released July 2026. State-of-the-art agentic coding with unusually low token usage, strong cybersecurity defense and science capabilities, and a 1.05M token context window.

OpenAI's most capable model, released September 2026. State-of-the-art in computer/browser use, software engineering, and cybersecurity, with a 1.05M context window. Uses opaque-recurrence reasoning whose chain of thought cannot be monitored — a point of controversy among safety researchers.

OpenAI's current image generation line, released September 2026 in two variants: Flare (fast default, ~50% lower latency than GPT Image 2) and Sunburst (precision editing). Six quality tiers from $0.006 to $0.211 per 1024x1024 image.

Google's high-efficiency multimodal model bringing near-Pro level coding and reasoning at Flash-tier cost. Optimized for agentic execution loops, supports text, image, video, audio, and PDF inputs with configurable thinking levels. Now a legacy model — superseded by Gemini 3.6, 3.7, and 3.8 Flash (July-September 2026); no shutdown announced.

Google's most intelligent Flash model, released September 2026 as the current default workhorse of the Gemini line. Configurable thinking levels (low/medium/high), 1M context, and a restricted Cyber variant scoring over 70% on vulnerability detection.

Google's current open-weight model family, released March 2026 under Apache 2.0. Sizes E2B, E4B, 26B, and 31B for on-device and self-hosted deployments; the Gemma family passed 1 billion downloads in August 2026.

Moonshot AI's flagship multimodal reasoning model, released July 2026 — the largest open-weight model to date at 2.8T total / 104B active parameters (MoE) with a 1M token context window and the Kimi Delta Attention architecture.

DeepSeek's current API workhorse, released September 2026. A 552B-parameter MoE with a novel Causal Encoder-Decoder architecture (8B active params for input, 16B for output), native vision, thinking and non-thinking modes, 1M context, and extreme cost-efficiency with off-peak discounts.

xAI's current flagship for coding and agentic tasks, released August 2026. 500K context, text and image input, reasoning effort levels from low to xhigh, and tiered pricing that doubles above 200K prompt tokens.

Meta Superintelligence Labs' proprietary flagship LLM, updated July 2026 (1.0 launched April 2026). Focused on multimodal reasoning, coding, tool use, visual chain-of-thought, and multi-agent orchestration. Unlike Llama, Muse is not open weights and is served through Meta AI consumer products.