ReferenceReference

☁️Frontier Models Reference

Frontier Models (Cloud API) Reference

Last verified: October 2026. Use this page to compare cloud costs against local deployment and to see what to escalate to when your local hardware falls short. Every price below is in USD per 1M tokens, as listed on each vendor’s pricing page in October 2026. Prices change often, so confirm on the vendor page before you commit: Anthropic, OpenAI, Google, xAI.


The Cloud Providers

If your local model is failing at a complex task, these are the flagship models you escalate the problem to.

Anthropic (Claude)

All four current models have a 1M token context window. Anthropic recommends starting with Opus 5.5 and moving up to Fable 5.1 when your evals on Opus 5.5 at higher effort still fall short.

  • Fable 5.1 (claude-fable-5-1): demanding reasoning and long-horizon agent work. $10 / $50.
  • Opus 5.5 (claude-opus-5-5): long-running agentic coding and knowledge work. $4 / $20.
  • Sonnet 5.5 (claude-sonnet-5-5): the speed and intelligence balance. $2 / $10.
  • Haiku 5.5 (claude-haiku-5-5): high-volume, latency-sensitive jobs like classification and routing. From $0.10 / $0.50.

OpenAI (GPT-6)

All four GPT-6 models have a 1.05M token context window and up to 128K output tokens.

  • GPT-6 Astra (gpt-6-astra): $10 / $50.
  • GPT-6.1 Sol (gpt-6.1-sol): $2 / $10.
  • GPT-6.1 Terra (gpt-6.1-terra): $1 / $5.
  • GPT-6 Luna (gpt-6-luna): $0.10 / $0.50.

Google (Gemini)

Gemini 4 Argon has a 1M token context window, and so does Gemini 3.8 Flash (1,048,576 input tokens). Google has scheduled price increases, shown in the table.

  • Gemini 4 Argon (gemini-4-argon): the current flagship. No free tier.
  • Gemini 3.8 Flash (gemini-3.8-flash): has a free API tier.
  • Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite): the cheapest current tier, with a free tier.

xAI (Grok)

Grok 4.7 (grok-4.7) has a 500K context window. Prompts of 200K tokens or more are billed at the higher rate for the whole request.


Pricing Comparison

ProviderModelInput ($/1M)Output ($/1M)Context window
AnthropicClaude Fable 5.1$10$501M
AnthropicClaude Opus 5.5$4$201M
AnthropicClaude Sonnet 5.5$2$101M
AnthropicClaude Haiku 5.5from $0.10from $0.501M
OpenAIGPT-6 Astra$10$501.05M
OpenAIGPT-6.1 Sol$2$101.05M
OpenAIGPT-6.1 Terra$1$51.05M
OpenAIGPT-6 Luna$0.10$0.501.05M
GoogleGemini 4 Argon$2 (becomes $4 on Dec 1, 2026)$10 (becomes $20)1M
GoogleGemini 3.8 Flash$0.75 (becomes $1.50 on Jan 1, 2027)$3.75 (becomes $7.50)1M
GoogleGemini 3.5 Flash-Lite$0.30$2.50not listed
xAIGrok 4.7$2 ($4 at 200K+ tokens)$6 ($12 at 200K+ tokens)500K

Batch APIs and prompt caching can cut these numbers a lot. Anthropic, for example, charges 50% less for batch requests. Check each vendor’s page for the details.

Mistral and DeepSeek also run APIs, but their current model prices were not confirmable at the time of writing, so they are left out. See Mistral’s pricing and DeepSeek’s pricing.


When to use Cloud vs. Local

We recommend Local-First, Cloud-Fallback.

Use Local Models when:

  • The data is highly sensitive, PII, or proprietary code.
  • You are doing high-volume/batch processing (e.g., summarizing 10,000 news articles).
  • The task is simple (RAG retrieval parsing, sentiment analysis, basic scripting).
  • You are offline.

Use Frontier APIs when:

  • You are working on complex, multi-file code refactoring.
  • You need deep, multi-step mathematical or scientific reasoning, which the top reasoning tiers (GPT-6 Astra, Claude Fable 5.1, Gemini 4 Argon) handle best.
  • You need to put a very large amount of context into one prompt. Claude, GPT-6 and Gemini all offer about 1M tokens.

Free Tiers for Trying It Out

If you don’t have the hardware for local models yet, Google lists a free API tier (with restrictive rate limits) for Gemini 3.8 Flash and Gemini 3.5 Flash-Lite. Gemini 4 Argon has no free tier. Try them in Google AI Studio.


For open-weight model alternatives you can run locally, see the Models Reference.

For routing between local and cloud autonomously, check out our Local + Cloud Module.