AI Models Catalog 2026

Claude Opus 4.7

Anthropic

The model GitHub Copilot uses. Best-in-class for nuanced reasoning, long-context analysis, and agentic workflows.

• Industry-leading code generation + tool-use
• Native computer-use + bash/file tools
• Strong refusal behavior — safer for enterprise

GPT-5

OpenAI

OpenAI's flagship. Strongest math/STEM reasoning + widest ecosystem of tools & integrations.

• Best-in-class math + competitive coding
• Native voice + Realtime API
• Function calling + Structured Outputs

Gemini 2.5 Pro

Google

Google's flagship. Unrivaled context length (2M tokens) — drop an entire codebase, get answers.

• Longest production context window (2M)
• Native video understanding
• Free tier via Google AI Studio

Grok 4

xAI

xAI's reasoning-first model with real-time X (Twitter) data access.

• Real-time web + X data
• Strong on STEM benchmarks
• API-first pricing

Claude Sonnet 4.5 (Code)

Anthropic

The default for Cursor, Windsurf, and GitHub Copilot's Agent mode. Reigning champion of agentic coding.

• #1 on SWE-bench Verified (real-world code tasks)
• Excellent tool-use + multi-step debugging
• Available in Cursor, Windsurf, Copilot, Aider

GPT-5 Codex

OpenAI

OpenAI's coding-tuned variant. Powers ChatGPT's Canvas + the Codex agent.

• Top-tier on HumanEval
• Built-in code interpreter
• Long-form refactors

DeepSeek-Coder V3

DeepSeek

Open-weight Chinese coding model — punches at frontier-class on benchmarks, free to self-host.

• Open-weight + huge context
• Strong on multi-file repo tasks
• Free to self-host

Codestral 25.06

Mistral

Mistral's fast, lightweight code completion model — built for IDE autocomplete latency.

• Optimized for FIM (fill-in-middle)
• Sub-100ms first-token latency
• 80+ languages

Llama 4 (70B Instruct)

Meta

Meta's flagship open-weight model. The de-facto baseline for self-hosted LLM stacks.

• Best 70B-class quality (open)
• Available on Bedrock, Together, Groq
• Permissive license

Qwen 2.5 72B

Alibaba

Alibaba's open-weight model — strong multilingual + math, Apache-2.0 license.

• Apache-2.0 (truly open)
• Excellent multilingual + math
• Many sizes from 0.5B to 72B

Mistral Large 2

Mistral

Mistral's flagship. Open-weight under research license; great agent + tool-use performance.

• Strong function-calling
• Available on Bedrock + Azure AI
• Top European model

Gemma 3 (27B)

Google

Google's open-weight family. Punchy quality per parameter — perfect for Ollama / single-GPU.

• Excellent quality-per-param
• Sizes from 1B to 27B
• First-class Ollama support

Phi-4

Microsoft

Microsoft's small-but-mighty 14B model — frontier-class reasoning at a fraction of the cost.

• Best-in-class reasoning at 14B
• MIT licensed
• Trivial to self-host

Claude 4 Vision

Anthropic

Best general-purpose vision LLM — reads diagrams, screenshots, whiteboards as well as humans.

• Excellent OCR + diagram understanding
• Vision-aware tool use
• Screenshot debugging

GPT-5 Vision

OpenAI

OpenAI's vision-enabled flagship. Strong on charts, math problems from photos, and image-to-code.

• Image-to-code (screenshot → JSX)
• Best for chart/graph reading
• Tight Code Interpreter integration

Llama 4 Vision

Meta

Best open-weight vision model. Self-hostable on a single H100 — full data control.

• Open weights + commercial use
• Strong on real-world photo Q&A
• Available on Bedrock

Flux 1.1 Pro

Black Forest Labs

The current SOTA in text-to-image. Beats Midjourney + DALL-E 3 on prompt adherence + photorealism.

• #1 on Artificial Analysis Image Arena
• Open-weight Schnell variant
• Available on Replicate, fal.ai

Stable Diffusion 3.5

Stability AI

Open-weight champion. Massive ComfyUI + Automatic1111 ecosystem — endless customization.

• Huge LoRA + ControlNet ecosystem
• Self-hostable on consumer GPUs
• Commercial license available

DALL-E 3

OpenAI

OpenAI's image model — strongest at typography + text rendering inside images.

• Best text-in-image rendering
• Native ChatGPT integration
• Strong prompt adherence

Midjourney v7

Midjourney

The artist's favorite. Unmatched aesthetic + style consistency for creative work.

• Best-in-class aesthetic quality
• Style reference + character consistency
• Web app now (no Discord needed)

Whisper Large v3

OpenAI

Open-weight speech-to-text. Industry standard for transcription — 99 languages, free to self-host.

• MIT licensed + self-hostable
• 99 languages including Hindi/Hinglish
• Best free transcription option

ElevenLabs v3

ElevenLabs

The most realistic AI voices on the market. Voice cloning + multilingual TTS at production quality.

• Sub-200ms streaming latency
• 32 languages incl. Hindi
• Voice cloning from 30 sec sample

Deepgram Nova-3

Deepgram

Lowest-latency speech-to-text API. Built for real-time voice agents.

• <300ms end-to-end latency
• Streaming + batch
• Speaker diarization

text-embedding-3-large

OpenAI

OpenAI's flagship embeddings — the safe default for RAG pipelines.

• Configurable output dims (256–3072)
• Top of MTEB leaderboard (proprietary)
• Cheap at scale ($0.13/M tokens)

Cohere Embed v4

Cohere

Best multilingual embeddings on the market. Native support for 100+ languages incl. Hindi.

• 100+ languages
• Multimodal (text + image)
• Free tier on dashboard

BGE-M3

BAAI

Top open-weight embedding model. Dense + sparse + multi-vector in one model. Free.

• MIT licensed
• Hybrid dense+sparse retrieval
• 100+ languages

nomic-embed-text-v1.5

Nomic

Small, fast, free embeddings — drop into Ollama in one command. Perfect for local RAG.

• Runs on CPU
• Apache-2.0
• Native Ollama support

Amazon Nova Pro

AWS

Amazon's own flagship FM. Cheap, fast, multimodal — only on Bedrock.

• AWS-native (IAM, VPC, PrivateLink)
• Multimodal incl. video
• Lowest $$/token of frontier models

Claude 4 (on Bedrock)

Anthropic / AWS

Claude with full AWS IAM + VPC + PrivateLink. The enterprise way to use Claude in India.

• Available in Mumbai region
• Provisioned throughput option
• Bedrock Guardrails + KMS

Llama 4 (on Bedrock)

Meta / AWS

Meta's open-weight Llama with managed scaling on Bedrock. No GPU ops needed.

• Fully managed — no GPU ops
• Knowledge Bases integration
• Agents for Bedrock

Mistral Large 2 (Bedrock)

Mistral / AWS

Mistral's flagship with AWS billing + IAM. Good for EU data residency too.

• Strong tool-use
• EU + US regions
• Per-token billing on Bedrock

LangGraph

LangChain

Stateful agent graphs from the LangChain team. The most-deployed agent framework in production today.

• Graph-based agent state machines
• Native human-in-the-loop
• LangSmith observability built-in

AWS Bedrock Agents

AWS

Fully-managed agents on AWS. Knowledge Bases, action groups, guardrails — no infrastructure to run.

• Zero-infrastructure agents
• Native IAM + KMS + VPC
• Knowledge Bases (RAG) built-in

OpenAI Agents SDK

OpenAI

OpenAI's official agent SDK. Tight integration with GPT-5, function calling, and Responses API.

• Native handoffs between agents
• Built-in tracing
• Python + TypeScript SDKs

Claude Agent SDK

Anthropic

Anthropic's official SDK for building autonomous Claude-powered agents with computer use + bash.

• Computer use + bash tool out-of-box
• File-system aware
• Powers Claude Code

Pydantic AI

Pydantic

Type-safe agent framework from the Pydantic team. FastAPI for the agent world — clean, opinionated, fast.

• Type-safe tool definitions
• Streaming + structured outputs
• Model-agnostic

Vercel AI SDK

Vercel

The fastest way to add streaming AI to a Next.js / React app. Used by ~40% of new AI startups.

• First-class React Server Components
• Streaming UI helpers
• 20+ provider plugins

CrewAI

CrewAI

Role-based multi-agent orchestration. Define agents like a real team — researcher, writer, reviewer.

• Role + goal + backstory primitives
• Sequential + hierarchical processes
• 30k+ GitHub stars

AutoGen v0.4

Microsoft

Microsoft's research-grade multi-agent framework. Event-driven, async, supports complex agent conversations.

• Async event-driven runtime
• Group chat patterns
• Strong code execution agents

LangGraph Swarm

LangChain

Swarm-style handoff agents on top of LangGraph. Inspired by OpenAI Swarm, production-hardened.

• Dynamic agent handoffs
• Persistent state across agents
• Time-travel debugging

AWS Multi-Agent Orchestrator

AWS

AWS's open-source multi-agent framework. Production-ready intent routing across Bedrock agents.

• Intent classifier routing
• Python + TypeScript
• Bedrock + Anthropic + custom

LangSmith

LangChain

The Datadog for agents. Trace every LLM call, debug step-by-step, evaluate prompts at scale.

• Full trace tree per agent run
• Datasets + evaluators built-in
• Free tier for solo devs

AgentOps

AgentOps.ai

Vendor-neutral agent monitoring. Track cost, latency, errors, and session replays across any framework.

• Framework-agnostic (CrewAI, AutoGen, LangChain, etc.)
• Session replay
• Cost + token tracking

Langfuse

Langfuse

Open-source LLM observability. Self-hostable, OpenTelemetry-native, vendor-agnostic.

• Self-hostable + cloud option
• OpenTelemetry-native
• Prompt management + datasets

Helicone

Helicone

Proxy-based LLM observability. One line of code change, full request logging + caching + cost analytics.

• Drop-in OpenAI proxy
• Built-in caching
• Custom properties for filtering

Arize Phoenix

Arize AI

Open-source LLM evals + tracing from the Arize team. Best-in-class evaluation framework.

• Strong RAG evals (faithfulness, relevance)
• OpenTelemetry-based
• Self-hostable

Pinecone

Pinecone

Managed serverless vector DB. The default for production RAG when you don't want to run infrastructure.

• Serverless + pay-per-query
• Hybrid (dense + sparse) search
• AWS / GCP / Azure

Qdrant

Qdrant

Rust-based open-source vector DB. Fastest single-node performance + Apache-2.0.

• Best raw throughput in benchmarks
• Self-hostable in Docker / K8s
• Rich filtering

Weaviate

Weaviate

Open-source vector DB with native multi-modal support. Strong for image + text RAG.

• Native multimodal (CLIP)
• GraphQL + REST APIs
• BYO-embeddings or auto-vectorize

pgvector

PostgreSQL

Vector search inside Postgres. Already running Postgres? You probably don't need a separate vector DB.

• No new infrastructure
• ACID transactions for vectors
• Available on RDS, Supabase, Neon

ChromaDB

Chroma

The easiest vector DB to start with. pip install, run in-process, perfect for prototypes.

• Embeddable + standalone modes
• Apache-2.0
• Best DX for learning

Mem0

Mem0

Memory layer for AI agents. Personalized agent recall across sessions — open-source.

• Long-term agent memory
• Vector + graph hybrid
• OpenAI / Anthropic / Bedrock support