Model Selection

Which models appear in your model selection is decided by your organization – not every environment offers all of the models described here. In most environments, gpt-6-luna is the preselected model for new chats.

Rule of thumb:

Fast & inexpensive → Luna / Mini / Nano / Flash / Haiku
Standard & reliable → large all-round models
Complex & critical → most powerful models
Hosting in Germany / open source → STACKIT models in CompanyGPT Sovereign or, if available, in your environment

Models marked as deprecated are listed under Legacy models, together with the model you should switch to.

OpenAI

The GPT-6 family

In CompanyGPT, GPT-6 is available in two tiers. They belong to the same generation and differ in how much reasoning depth, speed, and cost you get: Sol (the more capable tier) and Luna (fastest and most economical). The Sol tier is also available in a newer version, gpt-6.1-sol. Choose the tier based on the complexity of the task, not out of habit.

gpt-6.1-sol

For: demanding tasks where you want the newest version of the Sol tier
When: agentic coding, multi-step development tasks, complex analyses and workflows
Why: the updated version of gpt-6-sol – same tier, newer release

gpt-6-sol

For: demanding tasks with strong intelligence at fair cost
When: agentic coding, multi-step development tasks, complex workflows
Why: the more capable tier of the GPT-6 generation with careful validation

gpt-6-luna (default model)

For: fast answers at minimal cost
When: everyday tasks, high-volume applications, automated pipelines
Why: the most economical and lightweight model of the GPT-6 family – preselected for new chats in most environments

The GPT-5.6 family

GPT-5.6 is available in three tiers. They belong to the same generation and differ in how much reasoning depth, speed, and cost you get: Sol (maximum capability), Terra (balanced), and Luna (fastest and most economical). Choose the tier based on the complexity of the task, not out of habit.

GPT-5.6 Sol

For: the flagship tier with the highest capability
When: intricate software architecture, advanced programming, cybersecurity, deep multi-step scientific analysis, heavy creative strategy, oversight of other models — whenever precision outweighs cost
Why: the strongest reasoning of the 5.6 generation

GPT-5.6 Terra

For: the balanced mid-range tier for everyday professional and business workflows
When: standard business applications, document drafting, general assistant workflows
Why: the middle ground between speed and intelligence

GPT-5.6 Luna

For: the fastest and most cost-efficient tier
When: routine high-frequency automation, fast parsing, highly parallel agent tasks, and other high-volume or speed-sensitive work of lower complexity
Why: lowest latency and cost in the 5.6 generation

gpt-5.4

For: demanding tasks across many topics
When: coding, analyses, text work, and planning in everyday work
Why: strong all-round model of the GPT-5 generation; for the highest capability, choose GPT-5.6 Sol, gpt-6-sol or gpt-6.1-sol

gpt-5.1

For: deep thinking with high reliability
When: complex documents, data analyses, advanced text work
Why: strong GPT-5 generation performance optimized for practical use

gpt-5-mini

For: high quality at high speed
When: everyday programming tasks, structured research, logical iterations
Why: excellent balance of GPT-5 intelligence and fast processing time

gpt-5-nano

For: ultra-fast answers and simple assistance
When: quick questions, simple formatting, text corrections
Why: extremely resource-efficient model of the GPT-5 family with minimal latency

GPT Image 1.5

For: image generation and image editing
When: when images need to be created or edited
Why: OpenAI’s image model, processed in the EU data zone

GPT Image 2.5

For: image generation and image editing with the newer generation of OpenAI image models
When: when images need to be created or edited and your environment uses GPT Image 2.5
Why: newer generation of OpenAI image models; processing takes place globally and not in the EU data zone

You create images with the OpenAI Image Tools. Which image model is behind them is decided by your environment – in most environments it is GPT Image 1.5, in some GPT Image 2.5 or GPT Image 2. GPT Image 2 and GPT Image 2.5 run outside the EU data zone.

Google

The Gemini models are available if Google is enabled in your environment. The models of the Gemini 2.5 generation have been superseded, see Legacy models.

Gemini 4 Argon

For: demanding tasks with a lot of context
When: large documents and codebases (up to 1M tokens of context), complex analyses and multi-step reasoning
Why: the first model of the Gemini 4 generation, processed in an EU region; so far only available in some environments

Gemini 3.8 Flash / Gemini 3.7 Flash / Gemini 3.5 Flash

For: autonomous reasoning at high speed
When: complex workflows, deep analyses, demanding coding
Why: Google’s fast models with flexible thinking levels; 3.8 Flash is the newest Flash tier

Gemini 3.5 Flash-Lite / Gemini 3.1 Flash-Lite

For: maximum speed and cost efficiency
When: quick follow-up questions, simple bulk tasks, real-time translations
Why: extremely short response time at very low cost; 3.5 Flash-Lite is the newer tier

Gemini 3.1 Pro (preview)

For: particularly demanding analyses and complex reasoning
When: when the Flash models reach their limits, for example with very extensive or nested tasks
Why: Google’s more capable Pro tier; so far only available in some environments and as a preview, so its behavior and availability may still change

Gemini Image Tools / Nano Banana

For: image analysis, image generation, image editing
When: text-to-image generation, image editing with prompts (image + text), and composition of multiple images
Why: Google’s image models integrated into CompanyGPT; Nano Banana is processed in an EU region

In some environments, the Gemini Image Tools use the more capable Nano Banana Pro. It is processed globally and not in an EU region.

Anthropic

The Claude models are available if Anthropic is enabled in your environment. The current generation is Claude 5.5 with the Opus, Sonnet and Haiku tiers; it runs via the EU region of Google Vertex AI. The Claude 4.x models are deprecated, see Legacy models.

Claude Opus 5.5

For: the most demanding tasks, where quality comes before cost
When: complex analyses, extensive code changes, long multi-step agentic tasks
Why: the most capable Claude model of the current 5.5 generation; successor to Claude Opus 5 and the Opus 4.x models

Claude Sonnet 5.5

For: programming, complex text processing, and demanding all-round tasks (Recommended)
When: software development, code refactoring, agents, deep text comprehension
Why: the sweet spot of the series – high capability for everyday work at a considerably lower cost than Opus; successor to Claude Sonnet 5 and Sonnet 4.6

Claude Haiku 5.5

For: very fast processing at very low cost
When: filtering large data volumes, UI-based chatbots, simple to medium tasks at scale, subtasks of agents
Why: the fastest and most economical model of the 5.5 generation; successor to Claude Haiku 4.5

Claude Opus 5 / Claude Sonnet 5 (previous generation)

For: the most demanding tasks (Opus 5) or programming and all-round tasks (Sonnet 5)
When: if your environment does not offer the 5.5 generation yet
Why: the previous generation of Claude 5.5; once available, switch to Claude Opus 5.5 or Claude Sonnet 5.5

STACKIT

These open-source models run in the STACKIT Cloud in data centers in Germany and are particularly suited for workloads with high requirements for data sovereignty and internal compliance. They are available in CompanyGPT Sovereign and otherwise only in environments where STACKIT is enabled; the exact set of models differs from environment to environment. In addition, the search in companyRAG uses its own embedding models depending on the environment; they cannot be selected in the chat.

google/gemma-4-31B-it

For: versatile instruct tasks with good efficiency
When: internal assistants, knowledge work, structured text production
Why: newer Gemma generation as a mid-range option between small fast and large OSS models

Qwen/Qwen3.8-27B

For: versatile text tasks and strong multilingual understanding
When: creative writing, logical problem-solving, precise translations
Why: newer Qwen generation with a good balance of compact size and capability

openai/gpt-oss-120b

For: highest OSS quality for complex text tasks
When: when internally hosted top performance is required instead of maximum speed
Why: large open-source model for strong analytical and linguistic results

Qwen/Qwen3-VL-235B-A22B-Instruct-FP8

For: multimodal high-end analysis with image and text understanding
When: visual document analysis, complex image-text tasks, demanding inference
Why: very powerful VL model for deeper understanding of multimodal content

cortecs/Llama-3.3-70B-Instruct-FP8-Dynamic

For: more demanding generation and reasoning in the EU stack
When: more complex enterprise questions, longer responses, better level of detail
Why: the 70B class provides significantly more quality than small models while retaining OSS flexibility

openai/gpt-oss-20b

For: lightweight open-source text tasks
When: cost-sensitive internal workflows with controllable infrastructure
Why: compact OSS approach for solid quality with lower resource requirements

Legacy models

The following models are deprecated and are being superseded by newer generations. As long as your environment still offers them, you can keep using them; for new chats and agents, please choose the successor model named for each.

The entire GPT-4 family (gpt-4.1, gpt-4.1-mini, gpt-4o, gpt-4o-mini) will no longer be supported in the near future and will be removed from CompanyGPT; in some environments it is already unavailable. Please migrate to a current model: gpt-6-luna, GPT-5.6 Terra or gpt-5-mini as a replacement for gpt-4.1 and gpt-4o, and GPT-5.6 Luna or gpt-5-nano as a replacement for gpt-4.1-mini and gpt-4o-mini.

The Claude 4.x models (Claude Opus 4.5, 4.6, 4.7 and 4.8, Claude Sonnet 4.5 and 4.6, Claude Haiku 4.5) are being superseded by the Claude 5.5 generation: Claude Opus 5.5 replaces Opus 4.x, Claude Sonnet 5.5 replaces Sonnet 4.x, Claude Haiku 5.5 replaces Haiku 4.5.

The models of the Gemini 2.5 generation (Gemini 2.5 Pro, 2.5 Flash and 2.5 Flash-Lite) and older are no longer available in the chat. Their successors are Gemini 3.8 Flash for 2.5 Pro and 2.5 Flash, and Gemini 3.5 Flash-Lite for 2.5 Flash-Lite; where available, Gemini 3.1 Pro replaces 2.5 Pro. The Nano Banana image tool is not affected.

gpt-4.1

For: reliable problem-solving and standard contexts
When: text optimization, research, logical comparisons
Why: the proven, refined workhorse for demanding everyday tasks
Migrate to: gpt-6-luna / GPT-5.6 Terra / gpt-5-mini

gpt-4.1-mini

For: fast processing at low cost
When: structured text creation, data pre-sorting, simple conversations
Why: compact version of GPT-4.1, optimized for efficiency and speed
Migrate to: GPT-5.6 Luna / gpt-5-nano

gpt-4o

For: versatile multimodality and fluid interaction
When: image and speech processing, creative drafts, brainstorming
Why: the established flagship model for fast, multimedia tasks
Migrate to: gpt-6-luna / GPT-5.6 Terra / gpt-5-mini

gpt-4o-mini

For: high speed at minimal cost
When: simple chat assistants, fast filtering of large data volumes
Why: extremely inexpensive lightweight model with solid base intelligence
Migrate to: GPT-5.6 Luna / gpt-5-nano

Claude Opus 4.x (4.5, 4.6, 4.7, 4.8)

For: the most demanding tasks, where quality comes before cost
When: complex analyses, extensive code changes, multi-step agentic tasks
Why: the most capable Claude models of the 4.x generation
Migrate to: Claude Opus 5.5

Claude Sonnet 4.x (4.5, 4.6)

For: programming, complex text processing, and demanding all-round tasks
When: software development, code refactoring, deep text comprehension
Why: the previous sweet spot of the Claude series between capability and cost
Migrate to: Claude Sonnet 5.5

Claude Haiku 4.5

For: very fast processing with high logical precision
When: filtering large data volumes, UI-based chatbots, simple to medium tasks at scale
Why: fast and cost-efficient (less reasoning than Sonnet/Opus)
Migrate to: Claude Haiku 5.5

Quick decision-making aid

“I don’t know which model fits” → gpt-6-luna (default model)
“I just want a very good answer” → gpt-6-sol / gpt-6.1-sol / GPT-5.6 Terra / Claude Sonnet 5.5
“It should be as fast and inexpensive as possible” → gpt-6-luna / GPT-5.6 Luna / gpt-5-nano / Gemini 3.5 Flash-Lite / Claude Haiku 5.5
“I want to program / write code” → gpt-6.1-sol / gpt-6-sol / GPT-5.6 Sol / Claude Sonnet 5.5
“It’s complicated or extremely important” → GPT-5.6 Sol / gpt-6.1-sol / Claude Opus 5.5
“I’m running a lot of tasks in parallel” → gpt-6-luna / GPT-5.6 Luna / Claude Haiku 5.5
“Hosting in Germany is mandatory” → STACKIT models in CompanyGPT Sovereign or, if available, in your environment (e.g., Llama 3.3 70B Instruct)
“I need images” → OpenAI Image Tools (GPT Image 1.5 or GPT Image 2.5) / Gemini Image Tools (Nano Banana)
“I still use a GPT-4 or Claude 4.x model” → see Legacy models for the right successor model