Hoshi 1.0
Hoshi 1.0 is our proprietary frontier model, purpose-built for agentic workloads. It delivers frontier-tier performance on coding, structured output, and tool-use benchmarks at a fraction of the credits of leading alternatives.
Why Hoshi 1.0?
Hoshi 1.0 is designed from the ground up for the kinds of tasks EKB agents perform every day: reasoning over documents, calling tools, generating structured output, and following multi-step instructions. It is not a general-purpose chatbot — it is an agent engine.
Frontier-level quality at a fraction of the credits of GPT 5.5 or Sonnet 5.
Drop-in Chat Completions API — change three lines and keep your stack.
Standard request/response shape. Switch back or route around us anytime.
Pricing
| Tier | Credits |
|---|---|
| Input | 150 / 1M tokens |
| Cached Input | 10 / 1M tokens (~15× cheaper than fresh input) |
| Output | 300 / 1M tokens |
When to Use Hoshi vs. Frontier Models
Choosing the right model depends on the task. Use the guide below to decide when Hoshi 1.0 is the best fit and when you might reach for a frontier model like Sonnet 5 or GPT 5.5.
Use Hoshi 1.0 When…
| Scenario | Why Hoshi Wins |
|---|---|
| Tool calling & function use | Purpose-tuned for reliable, low-latency tool invocations — the core of any agent. |
| Structured output (JSON, tables) | Consistently produces valid, schema-conformant output without retries. |
| Multi-step reasoning | Excels at chaining actions, parsing responses, and deciding next steps. |
| High-volume batch tasks | At 150 credits / 1M input tokens, running thousands of agent turns is economical. |
| Knowledge Base retrieval | Optimized for RAG-style workflows — grounding answers in your documents. |
| Cost-sensitive production | Same quality tier as frontier models at a fraction of the credits. |
| Code generation & refactoring | Sweeps coding benchmarks; generates, edits, and reviews code reliably. |
Use a Frontier Model When…
| Scenario | Why a Frontier Model May Be Better |
|---|---|
| Ultra-long context (> 200K tokens) | Models like Claude Sonnet 5 or GPT 5.5 support very large context windows for massive document analysis. |
| Cutting-edge multimodal reasoning | When you need best-in-class vision, audio, or cross-modal understanding. |
| Highly creative / open-ended writing | Frontier models may produce more varied, stylistically rich prose for marketing or storytelling. |
| Regulatory or compliance constraints | Some organizations require a specific provider's model for data governance reasons. |
| Research-grade tasks | When you need the absolute latest capability on a novel, unstructured problem. |
For most EKB agent workloads — tool calling, structured output, RAG, and multi-step automation — Hoshi 1.0 is the recommended default. Use frontier models only when you have a specific reason to do so.
Quick Comparison
| Dimension | Hoshi 1.0 | Sonnet 5 | GPT 5.5 |
|---|---|---|---|
| Input cost | 150 credits / 1M | 2,000 credits / 1M | 5,000 credits / 1M |
| Output cost | 300 credits / 1M | 10,000 credits / 1M | 30,000 credits / 1M |
| Tool calling | ★★★★★ | ★★★★★ | ★★★★★ |
| Structured output | ★★★★★ | ★★★★☆ | ★★★★☆ |
| Coding | ★★★★★ | ★★★★★ | ★★★★★ |
| Creative writing | ★★★★☆ | ★★★★★ | ★★★★★ |
| Multimodal | ★★★☆☆ | ★★★★★ | ★★★★★ |
| Context window | 128K | 200K | 128K |
| OpenAI-compatible | ✅ | ❌ | ✅ |
Getting Started
Hoshi 1.0 is available in EKB as a predefined model. No API key configuration is required.
Select Hoshi in Your Agent
- Navigate to Agents
Go to Agents in the sidebar.
- Open or create an agent
Select an existing agent or create a new one.
- Go to the General tab
Open the General tab in the agent configuration.
- Select Hoshi 1.0
Find the Model drop-down selector and choose Hoshi 1.0 from the list.
FAQ
Is Hoshi 1.0 as good as Sonnet 5 or GPT 5.5?
For agentic tasks — tool calling, structured output, coding, and multi-step reasoning — Hoshi 1.0 performs at the same frontier tier. Frontier models may have an edge on highly creative writing, multimodal reasoning, and extremely long contexts.
What is cached input pricing?
When you send the same context repeatedly (e.g., the same system prompt or knowledge base retrieval), cached tokens are billed at 10 credits / 1M — roughly 15× cheaper than fresh input tokens.
Will Hoshi 1.0 be deprecated soon?
No. Hoshi 1.0 is our flagship model and will be actively maintained and improved. Future versions will be announced with clear migration timelines.
Learn More
- Model Pricing — compare all available models and their costs
- Agent Best Practices — tips for building effective agents