Model catalogue · Keera Gateway

The models behind your endpoint

Keera Gateway knows six model providers, four in Switzerland and two in the USA. Here is every model they serve, with its context window and its price in CHF per million tokens. You decide which team may use which model, and pay for usage at the provider's list price.

Start for free See pricing ➔

In Switzerland: stepping stone · Infomaniak · Phoeniqs · CSCS. In the USA: Anthropic · OpenAI.

Switzerland · open models

Models from stepping stone

stepping stone runs open models in its own data centres in Switzerland, and is one of the partners whose hardware Keera runs on.

Model	Context	InputCHF/MTok	OutputCHF/MTok
MiniMaxAI/MiniMax-M2.5	192k	1.94	9.70
NVIDIA/NVIDIA-Nemotron-3-Super-120B-A12B	128k	2.00	5.00
deepseek-ai/DeepSeek-V4-Flash-0731	512k	0.45	1.80
Qwen/Qwen3-Coder-Next	256k	0.34	1.70
zai-org/GLM-5.3-Flash	256k	0.30	1.50
openai/gpt-oss-120b	128k	0.40	1.60
Qwen/Qwen3.8-27B	256k	0.25	1.50
google/gemma-4-31B-it	160k	0.25	1.00
Qwen/Qwen3.5-35B-A3B-FP8	128k	0.17	1.00
apertus-ai/Apertus-v1.5-8B	128k	0.02	0.10
allenai/olmOCR-2-7B	8k	0.06	0.29
lightonai/LightOnOCR-2-1B	16k	0.02	0.06
opendatalab/MinerU2.5-2509-1.2B	16k	0.02	0.06

CHF per million tokens.

Switzerland · open models

Models from Infomaniak

Infomaniak runs open models in its own data centres in Switzerland, so prompts stay under Swiss law.

Model	Context	InputCHF/MTok	OutputCHF/MTok
Qwen/Qwen3.5-397B-A17B-FP8	200k	0.896	4.032
Qwen/Qwen3.5-122B-A10B-FP8	200k	0.448	3.584
moonshotai/Kimi-K2.6	256k	0.672	3.36
mistralai/Mistral-Small-4-119B-2603	256k	0.224	0.84
swiss-ai/Apertus-v1.5-70B	100k	0.784	2.80
google/gemma-4-31B-it	100k	0.224	0.448
mistralai/Ministral-3-14B-Instruct-2512	100k	0.336	0.448

CHF per million tokens.

Switzerland · open models

Models from Phoeniqs

Phoeniqs runs open models in its own data centres in Switzerland, and some of them read images too. The names are Phoeniqs's own: inference-glm5 is GLM-5.2 today, and a name stays when the model behind it gets a minor upgrade.

Model	Context	InputCHF/MTok	OutputCHF/MTok
inference-glm5	131k	1.077	3.386
inference-deepseek-v32	164k	0.6156	1.8469
inference-qwen3-vl-235b	128k	0.7003	2.00
inference-deepseek-v4-flash	1M	0.1462	0.3925
inference-glm5-flash	1M	0.123	0.41
inference-apertus-v15-70b	262k	0.6195	2.2201
inference-gpt-oss-120b	131k	0.1154	0.4617
inference-llama4-maverick	1.05M	0.2693	1.0773
inference-gemma4-31b	131k	0.118	0.325
inference-llama4-scout-17b	62k	0.1924	0.6387
inference-qwq-32b	33k	0.9234	0.9234
inference-qwen3-8b	41k	0.0269	0.1062
inference-granite-33-8b	33k	0.1539	0.1539
inference-mistral-v03-7b	33k	0.1539	0.1539
inference-granite-vision-2b	8k	0.077	0.077
inference-miner-u25	8k	0.38	0.23

CHF per million tokens.

Switzerland · open models

Models from CSCS

The Swiss National Supercomputing Centre runs open models on its own machines in Switzerland, Apertus among them, with open weights and open training data. The thinking variants reason before they answer and make no tool calls.

Model	Context	InputCHF/MTok	OutputCHF/MTok
moonshotai/Kimi-K2.7-Code	262k	1.30	6.05
zai-org/GLM-5.2	976k	6.36	27.31
nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16	262k	0.30	1.49
google/gemma-4-31B-it	262k	0.06	0.42
swiss-ai/Apertus-v1.5-70B	262k	0.06	0.40
swiss-ai/Apertus-v1.5-70B-thinking	262k	0.06	0.40
swiss-ai/Apertus-v1.5-8B	262k	0.01	0.04
swiss-ai/Apertus-v1.5-8B-thinking	262k	0.01	0.04
swiss-ai/Apertus-70B-Instruct-2509	64k	0.06	0.40
swiss-ai/Apertus-8B-Instruct-2509	33k	0.01	0.04

CHF per million tokens.

USA · proprietary models

Models from Anthropic

The Claude models run on Anthropic's API in the USA, and that is where the prompts go. Keera Gateway keeps them blocked until you open them per team and data class.

Model	Context	InputCHF/MTok	Cached inputCHF/MTok	OutputCHF/MTok
claude-opus-5-5	1M	3.68	0.184	18.40
claude-sonnet-5-5	1M	1.84	0.092	9.20
claude-haiku-5-5 *	1M	0.092	0.0092	0.46
claude-fable-5-1	1M	9.20	0.23	46.00
claude-sonnet-5	1M	1.84	0.184	9.20
claude-haiku-4-5	200k	0.92	0.092	4.60
claude-opus-5	1M	4.60	0.46	23.00
claude-opus-4-8	1M	4.60	0.46	23.00
claude-sonnet-4-6	1M	2.76	0.276	13.80

CHF per million tokens. Cached input is input the provider serves from its prompt cache.
* Above 100,000 tokens in the prompt, claude-haiku-5-5 costs five times as much.

USA · proprietary models

Models from OpenAI

The GPT models run on OpenAI's API in the USA. As with Anthropic, the prompts leave the country, and the models stay blocked until you open them.

Model	Context	InputCHF/MTok	Cached inputCHF/MTok	OutputCHF/MTok
gpt-6-astra *	1.05M	9.20	0.92	46.00
gpt-6.1-sol *	1.05M	1.84	0.092	9.20
gpt-6-sol *	1.05M	1.84	0.184	9.20
gpt-6-luna *	1.05M	0.092	0.0092	0.46
gpt-5.6-sol *	1.05M	3.68	0.368	18.40
gpt-5.6-terra *	1.05M	1.84	0.184	11.04
gpt-5.6-luna *	1.05M	0.184	0.0184	1.104
gpt-5.5 *	1.05M	4.60	0.46	27.60
gpt-5.4 *	1.05M	2.30	0.23	13.80
gpt-5.4-mini	400k	0.69	0.069	4.14
gpt-5.4-nano	400k	0.184	0.0184	1.15
gpt-5.2	400k	1.61	0.161	12.88
gpt-5.1	400k	1.15	0.115	9.20
gpt-4.1	1.05M	1.84	0.46	7.36
gpt-4.1-mini	1.05M	0.368	0.092	1.472
gpt-4o	128k	2.30	1.15	9.20
gpt-4o-mini	128k	0.138	0.069	0.552

CHF per million tokens. Cached input is input the provider serves from its prompt cache.
* Above 272,000 tokens in the prompt, input costs twice as much and output one and a half times as much.

Models that are not in the catalogue

Keera Engine runs any open model vLLM can serve, your own fine-tune included - on Swiss GPUs or in your cluster. We quote its price on request.

Any other endpoint that speaks the OpenAI API works too: another provider, or a server in your own data centre. You then enter its price yourself, so your budgets count it.

Get in touch