Model Registry

The business guide to LLM models

Search, filter, and compare a dated model reference for business. Use the tools to frame trade-offs, then verify current pricing, availability, and terms with each provider.

GuideReference snapshot, May 2026
Selection workspace

Compare models by the job, not the hype

The right model changes by workload. This guide frames model choice around cost, privacy, context length, and the level of judgment each task needs.

Budget

Smaller API model

Compare current provider rates

Privacy

Local or managed boundary

Match access to the data risk

Context

Task-sized window

Longer is not automatically better

Support

Instruction-following fit

Test tone, escalation, and cost

Business model stack

Recommended routing by workload

Balanced

Support replies

Claude 3 Haiku

$High fit

Document extraction

GPT-4o mini

$High fit

Internal search

Gemini Flash

$Medium fit

Private records

Llama 3.1

FixedHigh fit

Advisor output

Start with a hosted API for speed. Move sensitive or high-volume workflows to open-source once usage patterns are clear.

14 models
API

GPT-4o

OpenAI

Flagship multimodal model with strong creative writing, reliable reasoning, and massive developer ecosystem.

EnterpriseFast
128K$2.50 / $10.00per 1M tokens
API

Claude 3.5 Sonnet

Anthropic

Top-tier code generation, complex reasoning, and nuanced instruction following. Excels at multi-file tasks.

ReasoningCodeEnterprise
200K$3.00 / $15.00per 1M tokens
API

Gemini 1.5 Pro

Google

Google's high-capability model with a massive 1M token context window for processing entire codebases or documents.

FastEnterprise
1M$1.25 / $5.00per 1M tokens
API

GPT-4o mini

OpenAI

OpenAI's cost-optimized model. Great balance of quality and price for routine tasks at scale.

Fast
128K$0.15 / $0.60per 1M tokens
API

Claude 3 Haiku

Anthropic

Near-free pricing with best-in-class safety and instruction following. Fast response times for real-time chat.

Fast
200K$0.25 / $1.25per 1M tokens
API

DeepSeek-V3

DeepSeek

Extremely cost-effective with competitive quality. Specialized chain-of-thought reasoning architecture.

ReasoningFast
685B (37B active)128K$0.27 / $1.10per 1M tokens
API

Gemini 1.5 Flash

Google

Ultra-fast and ultra-cheap. 1M token context with Google's generous free tier.

Fast
1M$0.075 / $0.30per 1M tokens
Open Source

Llama 3.1 405B

Meta

Largest openly available dense model. Exceptional at long-form content with massive community ecosystem.

Open SourceEnterprise
405B128KSelf-host / Self-hostper 1M tokens
Open Source

Qwen3-235B

Alibaba Cloud

Massive MoE model with efficient routing. Strong multilingual support (29+ languages) and built-in tool calling.

Open SourceEnterprise
235B (22B active)128KSelf-host / Self-hostper 1M tokens
Open Source

DeepSeek-R1

DeepSeek

Specialized chain-of-thought reasoning. MIT license, fully permissive for commercial use.

Open SourceReasoning
671B (37B active)128KSelf-host / Self-hostper 1M tokens
Free

Qwen3-Coder

OpenRouter

Purpose-built for code generation and debugging. Free tier via OpenRouter, zero cost to start.

FreeCode
128KFree
Free

Kimi-K2.5

OpenRouter

Strong general-purpose reasoning at zero cost. Excellent at long-document analysis.

Free
128KFree
API

Mistral Large 2

Mistral AI

Near-GPT-4 quality on code and reasoning. Fluent in 12+ languages with native-level quality.

Open SourceMultilingual
123B128K$2.00 / $6.00per 1M tokens
Open Source

Mixtral 8x22B

Mistral AI

Mixture-of-Experts design keeps latency low. Strong coding and technical problem solving.

Open SourceCodeFast
176B (39B active)65KSelf-host / Self-hostper 1M tokens
Understanding your options

Open-source vs. API models

Two paths to the same destination. The right choice depends on your budget, your team, and how much control you need.

Open-source

Self-hosted models

Download and run on your own servers. Full control over data, customization, and cost.

Data privacy, Nothing leaves your servers.
Fixed costs, Pay for hardware, not per token.
Customizable, Fine-tune on your own data.
No vendor lock-in, Switch models anytime.
Best for

Teams with GPU infrastructure, strict data requirements, or high-volume workloads where per-token pricing becomes expensive.

API

Hosted API models

Send requests to a hosted service. Zero infrastructure, pay only for what you use.

Zero setup, API key and you are running.
Always updated, Provider handles model upgrades.
Scalable, Handles spikes automatically.
Lower barrier, No ML expertise needed.
Best for

Startups, small teams, and businesses that want fast results without managing infrastructure. Ideal for prototyping and moderate-volume use.

Local hosting calculator

Will this model run on your device?

Type in a device, choose an open-source model, and see whether the local setup has enough memory and speed headroom.

Quantization
Runs, with normal limits

Llama 3.1 8B

Llama 3.1 8B should run locally, but keep context and parallel users modest.

Estimated memory7.2GB / 11.5GB
Weights4.7GB
KV cache0.9GB
Runtime1.6GB
Speed margin+0.5

Good portable setup for smaller local models and single-user testing.

Use MLX or Ollama and close memory-heavy apps before running long context windows.

Better local fits
Side by side

Comparison table

Pricing per 1 million tokens. Scroll horizontally on mobile.

ModelProviderTypeInput/1MOutput/1MContextQualitySpeed
GPT-4oOpenAIAPI$2.50$10.00128KExcellentFast
Claude 3.5 SonnetAnthropicAPI$3.00$15.00200KExcellentFast
Gemini 1.5 ProGoogleAPI$1.25$5.001MVery GoodFast
DeepSeek-V3DeepSeekAPI$0.27$1.10128KVery GoodFast
GPT-4o miniOpenAIAPI$0.15$0.60128KGoodVery Fast
Claude 3 HaikuAnthropicAPI$0.25$1.25200KGoodVery Fast
Gemini 1.5 FlashGoogleAPI$0.075$0.301MGoodVery Fast
Qwen3-CoderOpenRouterFreeFreeFree128KGoodVaries
Kimi-K2.5OpenRouterFreeFreeFree128KGoodVaries
Llama 3.1 405BMetaOpen SourceSelf-hostSelf-host128KExcellentVaries
Mistral Large 2MistralOpen SourceSelf-hostSelf-host128KVery GoodVaries

Pricing reflects published API rates as of May 2026. Open-source costs depend on your hosting infrastructure.

Reference estimator

Token cost calculator

Compare a dated price reference against a representative workload, then verify current provider terms before making a buying decision.

Per task
<$0.01
Monthly
$75.00
Annual
$900.00
Tasks / mo
15,000

Estimates based on published API pricing as of April 2026. Actual costs vary with prompt complexity and caching. Free-tier models may have rate limits.

Planning tool

Deployment planning brief

Set the work, review boundary, and hosting reality first. The result is a practical evaluation path, not a live model recommendation.

Planning step 1 of 4

What needs the most care first?

FAQ

Frequently asked questions

Everything you need to know about choosing and using LLM models for your business.

A Large Language Model (LLM) is an AI system trained on text and other data to understand and generate language. For businesses, LLMs can help summarize documents, qualify leads, draft replies, write code, and support customer operations. The useful question is not whether a model can generate text, but whether it improves a specific workflow with the right controls and review.
API models like GPT-4o or Claude are hosted by the provider: you send requests and pay under their current terms. Open-weight models like Llama or Qwen can be run on infrastructure you control, but hardware, hosting, operations, licensing, and model availability still shape the cost and flexibility. The trade-off is that self-hosting requires technical expertise and capacity planning.
Free access can be useful for prototyping, but availability, rate limits, data terms, and model versions change often. Treat a free tier as a way to evaluate a narrow workflow, then confirm the provider terms and operational requirements before using it for customer-facing work.
LLM costs are usually measured in input and output tokens. Start with: tokens per task × tasks per month × the provider rate per million tokens, then add retrieval, hosting, monitoring, and human review where relevant. The calculator below is a planning estimate, not a provider quote.
For customer support, prioritize instruction-following, escalation behavior, cost, and a clear approval boundary. Test candidate models against real but safely handled conversations, including edge cases, before routing a customer-facing workflow automatically.
OpenRouter is a unified API gateway that routes requests to multiple AI model providers through a single endpoint. Some models are offered free during promotional periods or because providers subsidize access to gain market share. The free tier typically has rate limits and may not offer the same SLAs as paid plans.
Yes, and you should design for it. Most LLM APIs follow the OpenAI chat completions format, so switching from GPT-4o to Claude to an open-source model requires minimal code changes. Tools like LiteLLM, OpenRouter, and the Vercel AI SDK abstract provider differences behind a single interface.
We help teams evaluate the workflow, model options, source data, review boundary, and delivery approach before deciding what to build. The appropriate scope can include model evaluation, integration planning, automation design, or a controlled implementation, depending on the work and written engagement.
Need help choosing?

We build AI systems that actually work

Not sure which model fits your business? We help companies select, integrate, and optimize LLMs for real workflows, from chatbots to document processing to custom automation.

  • Free discovery call to scope your AI project
  • Fixed-price quote before any work starts
  • Working prototype in 2 to 4 weeks