# EcoHash — full model and pricing reference

> EcoHash provides RTX Pro 6000 GPU cloud, dedicated inference endpoints, and
> OpenAI-compatible model APIs. Canonical site: https://ecohash.com. API base URL:
> https://api.ecohash.com/v1. All prices below are pulled live from the billing
> database.

## GPU pricing
- RTX Pro 6000 (96GB): $1.89/GPU-hour

## GPU configurations and live availability
- 1x RTX Pro 6000 (rtx-pro-6000-1x-us): 16 vCPU, 64GB RAM, 96GB VRAM total, $1.89/hour — available now
- 2x RTX Pro 6000 (rtx-pro-6000-2x-us): 32 vCPU, 128GB RAM, 192GB VRAM total, $3.78/hour — available now
- 4x RTX Pro 6000 (rtx-pro-6000-4x-us): 64 vCPU, 256GB RAM, 384GB VRAM total, $7.56/hour — available now
- 8x RTX Pro 6000 (rtx-pro-6000-8x-us): 128 vCPU, 512GB RAM, 768GB VRAM total, $15.12/hour — limited

Machine-readable feed (public, no authentication):
https://api.ecohash.com/platform/gpu-availability
Availability measured at: 2026-09-10T13:19:28Z
Every configuration runs on a single machine. Billed per second at the
per-GPU-hour rate. Configurations shown as "on request" have no whole free
machine at this moment — contact https://ecohash.com/contact and capacity can usually be
freed.

## Storage
- [Storage](https://ecohash.com/storage): persistent storage attached to GPU instances and clusters, in the same region as the GPUs.
- Cloud Drive: Ceph block storage, ReadWriteOnce — attached to one instance at a time. Default mount /workspace/data-01.
- Shared Filesystem: CephFS, ReadWriteMany — mounted by many instances at once and by every replica of a cluster; also the place fine-tuning reads base weights from and dedicated inference loads models from. Default mount /workspace/shared-01.
- Sizes: 50, 100, 200, 300, 400 or 500 GB. Storage outlives the instance it was attached to and re-attaches to the next one. A folder can be exported to a download link (valid 3 days). Suspended storage stays readable for 5 days. Rates: https://ecohash.com/pricing#storage-pricing (live from /platform/storage-skus).

## Solutions (products built on EcoHash)
- AI Video Studio (EcoHash Studio): https://videos.ecohash.com — generative video (Wan 2.2, LTX, Seedance 2.0), talking-head avatars, voice cloning and TTS, captions, music and a browser timeline editor. Prepaid wallet, billed per second / picture / minute with an exact quote before each render. Details: https://ecohash.com/solutions/videos
- AI Coding (EcoHash Coding): https://coding.ecohash.com — flat monthly plans (Starter $9 / Pro $19 / Max $49) for coding tools such as Cline, Roo Code, Continue, Claude Code and Aider, on GLM-5.2, GLM-5.3, DeepSeek-V4-Flash, MiniMax-M3 and Kimi-K3. 1 credit = $0.01; hard spending ceiling. OpenAI-compatible base URL https://api.coding.ecohash.com/v1; Anthropic-shaped https://api.coding.ecohash.com/anthropic. Details: https://ecohash.com/solutions/coding
- Digital Employees (Octok): https://octok.com — Navigator plus specialist AI employees (marketing, support, localization, finance, legal) for global expansion; free plan, 10-minute Global Readiness Assessment. Runs on EcoHash inference. Details: https://ecohash.com/solutions/digital-employees

## Fine-tuning and datasets
- [Fine-Tuning](https://ecohash.com/fine-tuning): managed LoRA / QLoRA fine-tuning of open models. QLoRA is selected automatically when the base would not fit the chosen GPU in bf16. 1, 2 or 4 GPUs per job (data-parallel; the recipe means the same thing at any count).
- Base models: every self-hosted platform LLM (Qwen, Llama, Gemma families, including vision-language causal LMs), or bring your own via the Model Registry — a Hugging Face repo id (resolved and pinned to a commit SHA, up to 200 GiB of weights, architecture checked, trust_remote_code never enabled) or weights already on an EcoHash shared filesystem.
- Configuration: form fields plus optional pasted JSON. Parameters are checked against a strict allowlist (LoRA r / alpha / dropout / target_modules / bias / use_rslora / modules_to_save; training learning_rate / num_epochs / micro_batch_size / gradient_accumulation_steps / max_seq_length / warmup_steps / warmup_ratio / weight_decay / lr_scheduler_type / optim / max_grad_norm / neftune_noise_alpha); unknown keys are rejected by name, never silently ignored.
- Templates: save the launch form as a named template and reload it later; every account starts with Fast iteration, Balanced and Quality starters. Jobs record which template they launched from.
- Billing: per GPU-minute while training or evaluating, with a cost range shown before launch. Failed validation or synthesis is not charged. If the balance reaches zero the job pauses on a checkpoint and can be resumed.
- Output: perplexity and exact-match evaluation; the adapter is registered and attached at runtime to a LoRA-enabled inference instance (multi-LoRA, no restart) and called through https://api.ecohash.com/v1 with a model alias such as ft10:1.
- [Datasets](https://ecohash.com/datasets): first-class, reusable training datasets. Upload JSONL once (direct to storage with a progress bar), validation in 1–10 s, then train any number of jobs against it. Two formats: chat-format rows ({"messages":[...],"completion":"..."}) that train directly, or seed rows ({"input":"..."}) plus a skill prompt, synthesized with an open teacher model (DeepSeek R1, Qwen3, GLM-4.5, Kimi K2) using the customer's own key (KMS envelope-encrypted).
- Dataset validation: system prompt required, at least 100 examples, 64 KB per field, length checked against the base model's context; warnings for fewer than 1,000 examples, duplicate inputs and high length variance. Datasets are account-visible to teammates and carry a card (description, tags, structure stats). Storage is not billed; 50 GiB quota per account by default.

## Models (38)
### Llama-3.1-8B-Instruct
- URL: https://ecohash.com/models/llama-3.1-8b-instruct
- Model ID: llama-3.1-8b-instruct
- Category: Language
- Context: 128000 tokens
- Pricing: $0.10 in · $0.10 out / 1M
- About: The Meta Llama 3.1 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction tuned generative models in 8B, 70B and 405B sizes (text in/text out). The Llama 3.1 instruction tuned text only models (8B, 70B, 405B) are optimized for multilingual dialogue use cases and outperform many of the available open source and closed chat models on common industry benchmarks.

### Kokoro-82M
- URL: https://ecohash.com/models/kokoro-82m
- Model ID: kokoro-82m
- Category: Text-to-speech
- Pricing: $1.00 / 1M tok
- About: Fast, lightweight text-to-speech model

### Qwen3-ASR-1.7B
- URL: https://ecohash.com/models/qwen3-asr-1-7b
- Model ID: qwen3-asr-1-7b
- Category: Speech-to-text
- Pricing: $0.05 / 1M tok
- About: The Qwen3-ASR family includes Qwen3-ASR-1.7B and Qwen3-ASR-0.6B, which support language identification and ASR for 52 languages and dialects. Both leverage large-scale speech training data and the strong audio understanding capability of their foundation model, Qwen3-Omni. Experiments show that the 1.7B version achieves state-of-the-art performance among open-source ASR models and is competitive with the strongest proprietary commercial APIs. 

### Qwen3-TTS
- URL: https://ecohash.com/models/qwen3-tts
- Model ID: qwen3-tts
- Category: Text-to-speech
- Pricing: $2.00 / 1M tok
- About: Qwen3-TTS covers 10 major languages (Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian) as well as multiple dialectal voice profiles to meet global application needs. In addition, the models feature strong contextual understanding, enabling adaptive control of tone, speaking rate, and emotional expression based on instructions and text semantics, and they show markedly improved robustness to noisy input text.

### Jina-Embeddings-V3
- URL: https://ecohash.com/models/jina-embeddings-v3
- Model ID: jina-embeddings-v3
- Category: Embedding
- Pricing: $0.02 / 1M
- About: jina-embeddings-v3 is a multilingual multi-task text embedding model designed for a variety of NLP applications. Based on the Jina-XLM-RoBERTa architecture, this model supports Rotary Position Embeddings to handle long input sequences up to 8192 tokens. 

### Jina-Embeddings-V4
- URL: https://ecohash.com/models/jina-embeddings-v4
- Model ID: jina-embeddings-v4
- Category: Embedding
- Pricing: $0.12 / 1M
- About: jina-embeddings-v4 is a universal embedding model for multimodal and multilingual retrieval. The model is specially designed for complex document retrieval, including visually rich documents with charts, tables, and illustrations.

### BGE-Reranker-V2-M3
- URL: https://ecohash.com/models/bge-reranker-v2-m3
- Model ID: bge-reranker-v2-m3
- Category: Reranker
- Pricing: $0.03 / 1M
- About: bge-reranker-v2-m3，Different from embedding model, reranker uses question and document as input and directly output similarity instead of embedding. You can get a relevance score by inputting query and passage to the reranker. And the score can be mapped to a float value in [0,1] by sigmoid function.

### Qwen3.5-35B-A3B
- URL: https://ecohash.com/models/qwen3.5-35b-a3b
- Model ID: qwen3.5-35b-a3b
- Category: Vision
- Context: 32768 tokens
- Pricing: $0.40 in · $0.40 out / 1M
- About: Model Overview Type: Causal Language Model with Vision Encoder Training Stage: Pre-training & Post-training Language Model Number of Parameters: 35B in total and 3B activated Hidden Dimension: 2048 Token Embedding: 248320 (Padded) Number of Layers: 40 Hidden Layout: 10 × (3 × (Gated DeltaNet → MoE) → 1 × (Gated Attention → MoE)) Gated DeltaNet: Number of Linear Attention Heads: 32 for V and 16 for QK Head Dimension: 128 Gated Attention: Number of Attention Heads: 16 for Q and 2 for KV Head Dimension: 256 Rotary Position Embedding Dimension: 64 Mixture Of Experts Number of Experts: 256 Number of Activated Experts: 8 Routed + 1 Shared Expert Intermediate Dimension: 512 LM Output: 248320 (Padded) MTP: trained with multi-steps Context Length: 262,144 natively and extensible up to 1,010,000 tokens.

### Z-Image-Turbo
- URL: https://ecohash.com/models/z-image-turbo
- Model ID: z-image-turbo
- Category: Image
- Pricing: $10.00 / 1M tok
- About: Z-Image is a powerful and highly efficient image generation model family with 6B parameters. Currently there are four variants:  🚀 Z-Image-Turbo – A distilled version of Z-Image that matches or exceeds leading competitors with only 8 NFEs (Number of Function Evaluations). It offers ⚡️sub-second inference latency⚡️ on enterprise-grade H800 GPUs and fits comfortably within 16G VRAM consumer devices. It excels in photorealistic image generation, bilingual text rendering (English & Chinese), and robust instruction adherence.  🎨 Z-Image – The foundation model behind Z-Image-Turbo. Z-Image focuses on high-quality generation, rich aesthetics, strong diversity, and controllability, well-suited for creative generation, fine-tuning, and downstream development. It supports a wide range of artistic styles, effective negative prompting, and high diversity across identities, poses, compositions, and layouts.  🧱 Z-Image-Omni-Base – The versatile foundation model capable of both generation and editing tasks. By releasing this checkpoint, we aim to unlock the full potential for community-driven fine-tuning and custom development, providing the most "raw" and diverse starting point for the open-source community.  ✍️ Z-Image-Edit – A variant fine-tuned on Z-Image specifically for image editing tasks. It supports creative image-to-image generation with impressive instruction-following capabilities, allowing for precise edits based on natural language prompts.

### Seedance 2.0
- URL: https://ecohash.com/models/ecolink-video-gen-2.0
- Model ID: ecolink-video-gen-2.0
- Category: Video
- Pricing: from $2.45 / 1M, by output resolution and whether a reference video is supplied — see https://ecohash.com/pricing#video-pricing

### DeepSeek-V4-Flash
- URL: https://ecohash.com/models/DeepSeek-V4-Flash
- Model ID: DeepSeek-V4-Flash
- Category: Language
- Pricing: $0.49 in · $1.42 out / 1M
- About: DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens.

### DeepSeek-V4-Pro
- URL: https://ecohash.com/models/DeepSeek-V4-Pro
- Model ID: DeepSeek-V4-Pro
- Category: Language
- Pricing: $0.91 in · $2.72 out / 1M

### qwen3-embedding-0.6b
- URL: https://ecohash.com/models/qwen3-embedding-0.6b
- Model ID: qwen3-embedding-0.6b
- Category: Embedding
- Pricing: $0.12 / 1M
- About: Qwen3-Embedding-0.6B has the following features:  Model Type: Text Embedding Supported Languages: 100+ Languages Number of Parameters: 0.6B Context Length: 32k Embedding Dimension: Up to 1024, supports user-defined output dimensions ranging from 32 to 1024

### qwen3-omni-30b-a3b-instruct
- URL: https://ecohash.com/models/qwen3-omni-30b-a3b-instruct
- Model ID: qwen3-omni-30b-a3b-instruct
- Category: Vision
- Context: 32768 tokens
- Pricing: $0.40 in · $0.80 out / 1M

### Fun-ASR-Nano
- URL: https://ecohash.com/models/fun-asr-nano
- Model ID: fun-asr-nano
- Category: Speech-to-text
- Pricing: $0.05 / 1M tok
- About: LLM-Powered Speech Recognition — 31 Languages, Dialects & Accents  End-to-end ASR trained on tens of millions of hours of data. Supports Chinese (+ dialects), English, Japanese, Korean, French, German, Spanish, and 24 more languages.

### GLM-5.2
- URL: https://ecohash.com/models/GLM-5.2
- Model ID: GLM-5.2
- Category: Language
- Pricing: $1.00 in · $3.00 out / 1M
- About: GLM-5.2 marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a solid 1M-token context. GLM-5.2's new capabilities include:  Solid 1M Context: A solid 1M-token context that stably sustains long-horizon work Advanced Coding with Flexible Effort: Stronger coding capabilities with multiple thinking effort levels to balance performance and latency Improved Architecture: We propose IndexShare, which reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9× at a 1M context length. We also improve GLM-5.2’s MTP layer for speculative decoding, increasing the acceptance length by up to 20% Pure Open: An MIT open-source license — no regional limits, technical access without borders

### qwen3-vl-8b-instruct
- URL: https://ecohash.com/models/qwen3-vl-8b-instruct
- Model ID: qwen3-vl-8b-instruct
- Category: Vision
- Context: 32768 tokens
- Pricing: $0.15 in · $0.50 out / 1M
- About: Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.  This generation delivers comprehensive upgrades across the board: superior text understanding & generation, deeper visual perception & reasoning, extended context length, enhanced spatial and video dynamics comprehension, and stronger agent interaction capabilities.  Available in Dense and MoE architectures that scale from edge to cloud, with Instruct and reasoning‑enhanced Thinking editions for flexible, on‑demand deployment.  Key Enhancements: Visual Agent: Operates PC/mobile GUIs—recognizes elements, understands functions, invokes tools, completes tasks.  Visual Coding Boost: Generates Draw.io/HTML/CSS/JS from images/videos.  Advanced Spatial Perception: Judges object positions, viewpoints, and occlusions; provides stronger 2D grounding and enables 3D grounding for spatial reasoning and embodied AI.  Long Context & Video Understanding: Native 256K context, expandable to 1M; handles books and hours-long video with full recall and second-level indexing.  Enhanced Multimodal Reasoning: Excels in STEM/Math—causal analysis and logical, evidence-based answers.  Upgraded Visual Recognition: Broader, higher-quality pretraining is able to “recognize everything”—celebrities, anime, products, landmarks, flora/fauna, etc.  Expanded OCR: Supports 32 languages (up from 19); robust in low light, blur, and tilt; better with rare/ancient characters and jargon; improved long-document structure parsing.  Text Understanding on par with pure LLMs: Seamless text–vision fusion for lossless, unified comprehension.

### qwen3-coder-30b-a3b-instruct
- URL: https://ecohash.com/models/qwen3-coder-30b-a3b-instruct
- Model ID: qwen3-coder-30b-a3b-instruct
- Category: Language
- Context: 32768 tokens
- Pricing: $0.10 in · $0.30 out / 1M
- About: Qwen3-Coder-30B-A3B-Instruct has the following features:  Type: Causal Language Models Training Stage: Pretraining & Post-training Number of Parameters: 30.5B in total and 3.3B activated Number of Layers: 48 Number of Attention Heads (GQA): 32 for Q and 4 for KV Number of Experts: 128 Number of Activated Experts: 8 Context Length: 262,144 natively.

### gpt-oss-20b
- URL: https://ecohash.com/models/gpt-oss-20b
- Model ID: gpt-oss-20b
- Category: Language
- Context: 128000 tokens
- Pricing: $0.20 in · $0.28 out / 1M
- About: gpt-oss-20b, OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases. for lower latency, and local or specialized use cases (21B parameters with 3.6B active parameters)

### ViiTorVoice-NAR
- URL: https://ecohash.com/models/viitor-voice-nar
- Model ID: viitor-voice-nar
- Category: Voice cloning
- Pricing: $0.10 in · $0.50 out / 1M
- About: ViiTorVoice-NAR is a non-autoregressive speech generation model for voice cloning, local speech editing, and emotion / paralinguistic speech control. 

### Whisper-Large-V3-Turbo
- URL: https://ecohash.com/models/whisper-large-v3-turbo
- Model ID: whisper-large-v3-turbo
- Category: Speech-to-text
- Pricing: $0.10 / 1M tok
- About: Whisper large-v3-turbo is a finetuned version of a pruned Whisper large-v3. In other words, it's the exact same model, except that the number of decoding layers have reduced from 32 to 4. As a result, the model is way faster, at the expense of a minor quality degradation. 

### FLUX.2 Klein
- URL: https://ecohash.com/models/flux2-klein
- Model ID: flux2-klein
- Category: Image
- Pricing: $20.00 / 1M tok
- About: The FLUX.2 [klein] model family are our fastest image models to date. FLUX.2 [klein] unifies generation and editing in a single compact architecture, delivering state-of-the-art quality with end-to-end inference in as low as under a second. Built for applications that require real-time image generation without sacrificing quality.  FLUX.2 [klein] 9B is a 9 billion parameter rectified flow transformer capable of generating images from text descriptions and supports multi-reference editing capabilities.  Our flagship small model. Defines the Pareto frontier for quality vs. latency across text-to-image, single-reference editing, and multi-reference generation. Matches or exceeds models 5x its size—in under half a second. Built on a 9B flow model with 8B Qwen3 text embedder, step-distilled to 4 inference steps.

### Qwen-Image
- URL: https://ecohash.com/models/qwen-image
- Model ID: qwen-image
- Category: Image
- Pricing: $30.00 / 1M tok
- About: Qwen-Image, an image generation foundation model in the Qwen series that achieves significant advances in complex text rendering and precise image editing. Experiments show strong general capabilities in both image generation and editing, with exceptional performance in text rendering, especially for Chinese.

### Qwen3.5-27B
- URL: https://ecohash.com/models/qwen3.5-27b
- Model ID: qwen3.5-27b
- Category: Vision
- Context: 32768 tokens
- Pricing: $0.30 in · $0.60 out / 1M
- About: Qwen3.5 27B dense vision-language model. Text + image input, 32K context, FP8. Supports tool/function calling and reasoning.

### Qwen3.6-27B
- URL: https://ecohash.com/models/qwen3.6-27b
- Model ID: qwen3.6-27b
- Category: Vision
- Context: 32768 tokens
- Pricing: $0.30 in · $0.60 out / 1M
- About: Qwen3.6 27B dense vision-language model. Text + image input, 32K context, FP8. Supports tool/function calling and reasoning.

### Qwen3.6-35B-A3B
- URL: https://ecohash.com/models/qwen3.6-35b-a3b
- Model ID: qwen3.6-35b-a3b
- Category: Vision
- Context: 32768 tokens
- Pricing: $0.40 in · $0.40 out / 1M
- About: Qwen3.6 35B-A3B MoE vision-language model (35B total / 3B active). Text + image input, 32K context, FP8. Supports tool/function calling and reasoning.

### Wan2.2-T2V-A14B
- URL: https://ecohash.com/models/wan22-t2v-a14b
- Model ID: wan22-t2v-a14b
- Category: Video
- Pricing: from $0.05 / sec, by output resolution and whether a reference video is supplied — see https://ecohash.com/pricing#video-pricing

### Wan2.2-S2V-14B (Presenter / Talking Head)
- URL: https://ecohash.com/models/wan22-s2v-14b
- Model ID: wan22-s2v-14b
- Category: Video
- Pricing: from $0.05 / sec, by output resolution and whether a reference video is supplied — see https://ecohash.com/pricing#video-pricing
- About: Speech-to-video: animates a portrait to speak driving audio.

### Stable Audio Open 1.0
- URL: https://ecohash.com/models/stable-audio-open
- Model ID: stable-audio-open
- Category: Music & audio
- Pricing: $8.00 / 1M tok
- About: Text to audio — music beds and sound effects, 44.1kHz stereo, up to 47 seconds.

### pyannote Speaker Diarization 3.1
- URL: https://ecohash.com/models/pyannote-diarization-3-1
- Model ID: pyannote-diarization-3-1
- Category: Diarization
- Pricing: $4.00 / 1M tok
- About: Speaker diarization — detects who spoke when. Returns speaker turns with start/end times; pair with a transcription model for labelled transcripts.

### LTX 2.5
- URL: https://ecohash.com/models/ltx-2-5
- Model ID: ltx-2-5
- Category: Video
- Pricing: from $0.05 / sec, by output resolution and whether a reference video is supplied — see https://ecohash.com/pricing#video-pricing
- About: Text-to-video with synchronized audio. Generates a soundtrack alongside the picture and returns a single MP4 with both tracks.

### Qwen3.8-27B
- URL: https://ecohash.com/models/qwen3.8-27b
- Model ID: qwen3.8-27b
- Category: Vision
- Context: 32768 tokens
- Pricing: $0.30 in · $0.60 out / 1M
- About: Qwen3.8 27B dense vision-language model. Text + image input, 32K context, FP8. Supports tool/function calling and reasoning.

### Kimi-K3
- URL: https://ecohash.com/models/Kimi-K3
- Model ID: Kimi-K3
- Category: Language
- Pricing: $2.00 in · $12.00 out / 1M

### GLM-5.3
- URL: https://ecohash.com/models/GLM-5.3
- Model ID: GLM-5.3
- Category: Language
- Pricing: $1.25 in · $4.30 out / 1M

### MiniMax-M3
- URL: https://ecohash.com/models/MiniMax-M3
- Model ID: MiniMax-M3
- Category: Language
- Pricing: $0.20 in · $0.90 out / 1M

### qwen3.8-max
- URL: https://ecohash.com/models/qwen3.8-max
- Model ID: qwen3.8-max
- Category: Language
- Pricing: $1.90 in · $5.70 out / 1M

### Wan 3.0
- URL: https://ecohash.com/models/wan3.0-video
- Model ID: wan3.0-video
- Category: Video
- Pricing: from $0.06 / sec, by output resolution and whether a reference video is supplied — see https://ecohash.com/pricing#video-pricing

### wan3.0-video-prime
- URL: https://ecohash.com/models/wan3.0-video-prime
- Model ID: wan3.0-video-prime
- Category: Video
- Pricing: from $0.07 / sec, by output resolution and whether a reference video is supplied — see https://ecohash.com/pricing#video-pricing

## Blog (7)
- [How to deploy and tune Qwen3.8-27B on one RTX Pro 6000](https://ecohash.com/blog/deploy-qwen38-27b-rtx-pro-6000): Qwen3.8-27B is 27B dense in FP8, so one RTX Pro 6000 holds it with room for a large KV
cache. The serving config EcoHash runs in production, the five settings whose obvious
alternative fails, and the one flag that doubled what a single card serves.
- [Run your Retell AI agent on an open model: the custom LLM bridge](https://ecohash.com/blog/run-your-retell-ai-agent-on-an-open-model-the-custom-llm-bridge): How to connect a Retell AI voice agent to EcoHash's OpenAI-compatible API with a small WebSocket bridge, what it does to the per-minute LLM bill, and what stays on Retell.
- [Give your Vapi voice agent a cheaper voice: Kokoro TTS via custom-voice](https://ecohash.com/blog/vapi-custom-tts-kokoro): How to plug EcoHash's Kokoro TTS into a Vapi voice agent as a custom voice: a small adapter server, the assistant config, a local test, and what it costs per minute compared to ElevenLabs.
- [Build a voice agent with Whisper, Kokoro, and an OpenAI-compatible API](https://ecohash.com/blog/kokoro-whisper-voice-agent): How to build a speech-to-text to LLM to text-to-speech voice agent on EcoHash using Whisper, a chat model, and Kokoro, all through one OpenAI-compatible API key.
- [Qwen3 Coder 30B on RTX Pro 6000: API, dedicated endpoint, or GPU workspace?](https://ecohash.com/blog/qwen3-coder-30b-rtx-pro-6000): Three ways to run Qwen3 Coder 30B on EcoHash — the shared OpenAI-compatible API, a dedicated endpoint, or an RTX Pro 6000 GPU workspace — and how to choose between them.
- [Best models to run on RTX Pro 6000 96GB](https://ecohash.com/blog/best-models-for-rtx-pro-6000): A task-by-task guide to choosing open models for a 96GB RTX Pro 6000 on EcoHash: coding, general chat, embeddings and rerankers for RAG, and speech models, all through an OpenAI-compatible API.
- [What makes RTX Pro 6000 96GB a good GPU for AI inference?](https://ecohash.com/blog/rtx-pro-6000-gpu-advantages): How the NVIDIA RTX Pro 6000 Blackwell Server Edition with 96GB VRAM fits open-model inference on EcoHash, which model sizes it suits, how it compares to other GPUs, and how to access it.