Deploy Agent
Docs/AI Models/Supported Models
Reference

AI Models

Lobstack supports 30+ AI models across every major provider. Switch models anytime from your dashboard — models are grouped by provider with badges for quick comparison.

Meet Lobstack 1.2 — our flagship model

Don't want to pick? Choose Lobstack 1.2, our flagship managed model. Behind one name, the Lobstack Gateway uses proprietary Token Intelligence to route every message in real time to the best engine for the job. With Use Lobstack AI there are no API keys to manage — usage is included in your plan. Prefer a specific engine? Pick any of the 30+ below.

Current Generation (July 2026)#

ModelProviderTierContext WindowNotes
Lobstack 1.2 (flagship)LobstackManagedToken Intelligence routes to the best engine per message. No API key needed.
Claude Opus 5AnthropicFlagship1MMost capable — complex reasoning, analysis, and agentic work
GPT-5.6 SolOpenAIFlagship1MOpenAI's flagship — state-of-the-art coding and reasoning
DeepSeek V4 ProDeepSeekFlagship1MFlagship-class reasoning at a fraction of the price
Gemini 3.1 ProGooglePremium1MStrong multimodal reasoning with long context
GPT-5.6 TerraOpenAIPremium1MBalanced GPT-5.6 variant for demanding everyday work
Grok 4.5xAIPremium500KReal-time knowledge, strong for current events
Qwen 3.7 MaxAlibabaPremium1MStrong multilingual, excellent for Asian-language tasks
Mistral Medium 3.5MistralPremium128KEfficient European model, multilingual
Claude Sonnet 5AnthropicStandard1MBest balance of capability and speed — recommended default
Gemini 3.5 FlashGoogleStandard1MVery fast, great for high-volume tasks
Qwen 3.7 PlusAlibabaStandard1MCost-effective general-purpose, fast inference
DeepSeek V4 FlashDeepSeekStandard128KExtremely cost-effective, strong coding and math
Claude Haiku 4.5AnthropicSmall200KFast and light for well-defined tasks
GPT-5.6 LunaOpenAISmall1MFast, cost-efficient GPT-5.6 for simpler tasks
Grok 4.1 FastxAINano2MUltra-fast and cheap with a huge context window

Previous Generation#

These models remain available for existing agent configurations. New agents should prefer the current generation above.

ModelProviderTierContext WindowNotes
Claude Opus 4.6AnthropicFlagship1MPrevious Anthropic flagship
GPT-5.2OpenAIFlagship256KPrevious OpenAI flagship
GPT-5OpenAIPremium256KAdvanced coding and complex reasoning
Mistral LargeMistralPremium128KStrong European model, multilingual
Qwen MaxAlibabaPremium32KStrong multilingual
DeepSeek R1DeepSeekPremium1MRetired by DeepSeek — automatically routes to DeepSeek V4 Pro
Claude Sonnet 4.5AnthropicStandard200KPrevious-generation Sonnet
GPT-4oOpenAIStandard128KStrong general-purpose, multimodal
Gemini 2.5 ProGoogleStandard1MStrong reasoning with large context window
Llama 3.3 70BMeta (via Groq)Standard128KOpen-source, ultra-fast inference via Groq
Grok 3xAIStandard131KPrevious-generation Grok
GPT-5 MiniOpenAISmall256KFast, cost-efficient for well-defined tasks
DeepSeek V3DeepSeekSmall128KRetired by DeepSeek — automatically routes to DeepSeek V4 Flash
Qwen PlusAlibabaSmall128KCost-effective general-purpose
GPT-5 NanoOpenAINano128KUltra-fast, cheapest for summarization and classification
Mistral SmallMistralNano128KFast and lightweight, great for high-volume tasks
💡

DeepSeek endpoint migration (July 2026)

DeepSeek retired its legacy chat and reasoner API endpoints on 2026-07-24. Nothing to do on your side: agents configured with the old DeepSeek V3 or DeepSeek R1 keys are automatically routed to DeepSeek V4 Flash and DeepSeek V4 Pro respectively.

Switching Models#

To change your AI model:

  1. Open the Dashboard and navigate to Settings
  2. Click Change model under the AI Model section
  3. Browse models grouped by provider — each model shows badges like Latest, Fast, Cheap, or Free to help you compare
  4. Select your model and click Save & Rebuild

The rebuild takes 2-4 minutes. Your chat history and memories are preserved — only the VM is recreated with the new model configuration.

💡

Starter plan restrictions

The Starter plan ($29/mo) does not include access to flagship and premium models such as Claude Opus 5, GPT-5.6 Sol, or Gemini 3.1 Pro. Upgrade to Pro ($59/mo) or higher to use all models.

Self-Hosted Models (Ollama)#

Ollama models run directly on your agent's VM — no external API key required. This means zero per-message cost and complete data privacy. The trade-off is that inference speed depends on your server tier's hardware. Performance or Enterprise plans are recommended for self-hosted models.

Bring Your Own API Key#

You can provide your own API key for any supported provider — Anthropic, OpenAI, Google, DeepSeek, Alibaba Cloud (Qwen), Mistral, Groq (Llama), xAI (Grok), or Moonshot. When configured, your agent will use your key instead of the Lobstack default. This gives you full control over billing and rate limits. The Settings panel automatically detects your model's provider and shows the correct key format.

To add your API key:

  1. Go to Settings in your dashboard
  2. Click Add API Key
  3. Paste your key and click Save & Rebuild

Key security

Your API key is sent securely to your VM and stored in the agent config on your server. It is never stored in the Lobstack database in plain text. You can clear or change it at any time.

Message Overage Pricing#

If you exceed your plan's monthly message limit, overage charges apply per message based on the model:

ModelProviderOverage / Message
Claude Opus 5Anthropic$0.05
GPT-5.6 SolOpenAI$0.04
DeepSeek V4 ProDeepSeek$0.01
Gemini 3.1 ProGoogle$0.02
GPT-5.6 TerraOpenAI$0.03
Grok 4.5xAI$0.02
Qwen 3.7 MaxAlibaba$0.01
Mistral Medium 3.5Mistral$0.015
Claude Sonnet 5Anthropic$0.02
Gemini 3.5 FlashGoogle$0.015
Qwen 3.7 PlusAlibaba$0.005
DeepSeek V4 FlashDeepSeek$0.005
Claude Haiku 4.5Anthropic$0.01
GPT-5.6 LunaOpenAI$0.015
Grok 4.1 FastxAI$0.005
Previous-generation modelsVarious$0.005 – $0.05 (unchanged)

Model Routing#

The Lobstack Agent Runtime automatically routes your selected model to the correct provider API. Internally, models are mapped to provider-prefixed identifiers (e.g., anthropic/claude-opus-5). This happens transparently — you just select the model name in your dashboard.