Model catalog

All platform models are called at https://api.ecohash.com/v1/... with your API key. Pass the model ID exactly as shown below in the model field of your request.

note

This page lists the models available at the time of writing. For an always-current, machine-readable snapshot — including live pricing and context lengths — call GET https://api.ecohash.com/platform/models. The model_id field in that response is the canonical ID to send.

Chat & reasoning (LLMs)

Called via POST /v1/chat/completions.

Model ID	Best for
llama-3.1-8b-instruct	Everyday chat, summarization, extraction — fast and inexpensive
qwen2.5-7b-instruct	Lightweight general chat
gpt-oss-20b	Open-weight reasoning and agentic tasks
qwen3-coder-30b-a3b-instruct	Code generation and tool-calling
Qwen3-235B-A22B	Large mixture-of-experts model for demanding reasoning
GLM-5.2	Long-horizon tasks and agentic workflows (large context)
DeepSeek-V4-Flash	High-throughput reasoning with very long context

Vision + language

Multimodal chat — send images alongside text in the messages array. Called via POST /v1/chat/completions.

Model ID	Best for
qwen3-vl-8b-instruct	Image understanding, OCR-style tasks, visual Q&A
gemma-4-31b-it	Higher-quality multimodal reasoning
qwen3.5-35b-a3b	Vision + language with a larger reasoning model
Kimi-K2.6	Long-horizon multimodal agentic work

See Chat completions for how to attach images.

Embeddings

Called via POST /v1/embeddings. Return dense float vectors for retrieval and semantic search.

Model ID	Notes
jina-embeddings-v3	Multilingual, multi-task text embeddings
jina-embeddings-v4	Multimodal / multilingual retrieval embeddings
qwen3-embedding-0.6b	Small, fast — 100+ languages
qwen3-embedding-4b	Mid-size embeddings
qwen3-embedding-8b	Largest, highest-quality embeddings

See Embeddings.

Reranker

Called via POST /v1/rerank. Given a query and candidate documents, returns a relevance score per document.

Model ID	Notes
bge-reranker-v2-m3	Multilingual cross-encoder reranker

See Reranker.

Image generation

Called via POST /v1/images/generations (and POST /v1/images/edits where supported).

Model ID	Notes
flux2-klein	Fast unified generation + editing
qwen-image	Strong text rendering and precise editing
z-image-turbo	Distilled, low-step turbo model for quick generations

See Image generation.

Speech-to-text (transcription)

Called via POST /v1/audio/transcriptions.

Model ID	Notes
whisper-large-v3	High-accuracy multilingual transcription
whisper-large-v3-turbo	Faster variant of Whisper Large V3
qwen3-asr-1-7b	Multilingual ASR with language identification
fun-asr-nano	Compact ASR across many languages, dialects, and accents

See Audio transcription.

Text-to-speech

Called via POST /v1/audio/speech. Supports streaming.

Model ID	Notes
kokoro-82m	Fast, lightweight TTS with a range of voices
qwen3-tts	Multilingual TTS (10+ languages and dialects)
chatterbox	Multilingual general-purpose TTS

See Text-to-speech.

Voice cloning & speech editing

A single model backs both, using dedicated multipart endpoints.

Model ID	Endpoints	Notes
viitor-voice-nar	POST /v1/audio/voice-clone, POST /v1/audio/text-local-edit	Clone a voice from a short reference clip, or edit words in existing audio while preserving the voice

See Voice cloning and Speech editing.

Video generation

Asynchronous — submit a job, then poll for the result. Called via POST /v1/video/generations.

Model ID	Notes
ecolink-video-gen-2.0	Text-, image-, and audio-conditioned video generation

Video is billed on real usage per completed job — see Video generation.

Availability & routing

Models are served from multiple regions. Actual routing depends on real-time load and health, so a request may be served from a different region than the one closest to you — this is transparent to your code.

• If a model is temporarily unavailable everywhere (rare), you get 503 Service Unavailable; retry after a short backoff.
• If you get 404 Not Found, the model ID isn't recognized — check the spelling against GET /platform/models.