Model Cards
Full GreenPT model catalog (chat, coding, embeddings, speech-to-text, and reranking) with pricing per model.
Full GreenPT model catalog. Filter by category and compare pricing. Token prices are in EUR per 1M tokens; speech-to-text models are billed per hour of audio. To list models programmatically, see the Models API.
Catalog
kimi-k3Moonshot AINewMoonshot Kimi K3, an open-weight 2.8T-parameter native multimodal agentic model with hybrid reasoning for long-horizon coding, knowledge work, and reasoning.
glm-5.2z.aiNewGreenPT Code flagship coding model with strong reasoning, agentic tool-use, and multi-file software engineering performance. See the GreenPT Code guide for IDE setup. Also available as output-compression variants (glm-5.2-caveman, -honey, -ponytail) at the same price.
deepseek-v4-flash-0731DeepSeekNewDeepSeek V4 Flash, 2026-07-31 snapshot. A fast, low-cost reasoning model with a 1M-token context window and tool use. Reasoning is always on, so allow enough output budget for it.
minimax-m2.5MiniMaxMiniMax reasoning model optimized for agentic coding and long-horizon workflows.
kimi-k2.6Moonshot AIMoonshot Kimi K2.6 reasoning model with strong agentic coding and multimodal input.
kimi-k2.7-codeMoonshot AIMoonshot Kimi K2.7 Code, tuned for software engineering and agentic coding tasks.
gemma4GreenPTGreenPT-branded next-generation chat model with long-context multimodal reasoning. Recommended default for chat completions.
green-rGreenPTGreenPT-branded reasoning model for advanced analysis, writing, and content generation.
green-r-rawGreenPTDirect access to the same reasoning stack as green-r without the GreenPT system prompt.
green-lGreenPTGreenPT-branded chat model tuned for multilingual writing, image understanding, and Dutch grammar guardrails.
green-l-rawGreenPTDirect access to the same GreenPT-backed model as green-l without the built-in system prompt.
qwen3.5-397b-a17bQwenState-of-the-art Qwen model optimized for code generation, agentic tasks, and logical reasoning.
mistral-small-3.2-24b-instruct-2506MistralGeneral-purpose model with balanced cost, function calling, and multimodal chat support.
gpt-oss-120bOpenAILarge open model with vision, function calling, and long-context reasoning support.
qwen3-235b-a22b-instruct-2507QwenHigh-context multilingual reasoning model with a 250k token window.
qwen3-coder-30b-a3b-instructQwenQwen coding model built for generation, completion, and debugging workflows.
llama-3.3-70b-instructMetaMeta instruction-following model with strong multilingual support for general chat use cases.
qwen3.6-35b-a3bQwenFrontier small Qwen model optimized for agentic and reasoning tasks across many languages.
mistral-medium-3.5-128bMistralUnified Mistral model with strong instruct, reasoning, coding, and vision performance.
green-embeddingGreenPTDense multilingual embedding model for semantic search, retrieval, and RAG pipelines. Supports Matryoshka Representation Learning, so output dimensions are configurable from 32 up to 2560.
qwen3-embedding-8bQwenState-of-the-art multilingual embedding model for semantic search, retrieval, and RAG pipelines. Supports Matryoshka Representation Learning, so output dimensions are configurable from 32 up to 4096.
bge-multilingual-gemma2BAAICompact multilingual embedding model, a cost-efficient option for large corpora where an 8k input window is enough.
green-sGreenPT50% off until 30 SepSpeech-to-text model for pre-recorded and live transcription.
green-s-proGreenPTPro50% off until 30 SepAdvanced speech-to-text model with multilingual transcription options.
green-rerankGreenPTCross-encoder reranking model that scores and reorders search or RAG results by relevance to a query.
Deprecated model ids
A pending id keeps answering and is served by its replacement until the listed date. A removed id is no longer listed by GET /v1/models from that date, and requests to it return 400 Unsupported model. A directGET /v1/models/{id} lookup returns 404.
devstral-2-123b-instruct-2512MistralDeprecation pendingserved by glm-5.2 from 1 September 2026gemma-3-27b-itGoogleDeprecation pendingserved by gemma4 from 1 September 2026holo2-30b-a3bH CompanyDeprecation pendingserved by qwen3.6-35b-a3b from 1 September 2026qwen3-coder-30b-a3b-instructQwenDeprecation pendingserved by deepseek-v4-flash-0731 from 1 October 2026pixtral-12b-2409MistralDeprecation pendingserved by mistral-small-3.2-24b-instruct-2506 from 1 October 2026deepseek-r1-distill-llama-70bDeepSeekDeprecatedremoved on 18 August 2026 · use deepseek-v4-flash-0731devstral-small-2505MistralDeprecatedremoved on 18 August 2026 · use deepseek-v4-flash-0731llama-3.1-8b-instructMetaDeprecatedremoved on 18 August 2026 · use deepseek-v4-flash-0731mistral-nemo-instruct-2407MistralDeprecatedremoved on 18 August 2026 · use deepseek-v4-flash-0731voxtral-small-24b-2507MistralDeprecatedremoved on 18 August 2026 · use green-s-proBuilt-in system prompts
Some ids carry a system prompt of their own, so the id you send decides which rules the model gets on top of yours:
- Compression models (
glm-5.2-caveman,-honeyand-ponytail, each also in-liteand-ultra) carry an output-compression ruleset: the same model and price per token asglm-5.2, with a shorter answer. See Compression models. - GreenPT models (
green-l,green-r) carry GreenPT's Sustainability system prompt, optimized for our use cases; their raw twins (green-l-raw,green-r-raw) carry none, so yours is the only system prompt sent upstream. - Every other id in the catalog carries no system prompt.
When a model carries one and you send one too, both go upstream as a single
system message: yours first, the model's rules last. Picking the id is an
explicit opt-in to that style, so on a conflict the model's rules win over a
generic prompt your client happens to send. A developer message counts as a
system message here and merges the same way.
Catalog Notes
All models are served through the GreenPT API proxy with an OpenAI-compatible interface. Prices are in EUR per million tokens unless noted otherwise, and API pricing applies to API usage only. For how we measure the environmental footprint of a request, see Carbon calculations.