Using Claude Code with InferenceHub — all models with one key, $10/mo.
Start free: https://app.inferencehub.tech/portal#/agents

Using Zcode with InferenceHub $10 plan for daily development work

https://zcode.z.ai/en

Created a group, please feel free to join and ask any questions you might have.
https://t.me/inferencehub_group

🖥️ InferenceHub Desktop is here

The full InferenceHub chat, now a native app for macOS and Windows — every model on one key: real Claude, GPT, and open-weight models, plus web search, deep research, and project files, outside the browser.

Download → https://inferencehub.tech/download
• macOS (Apple Silicon)
• Windows 10/11 (x64)

Sign in with your InferenceHub account — $1 free credit to start.

One honest heads-up: this is an early unsigned preview, so your OS will warn on first launch:
• Windows: More info → Run anyway
• macOS: right-click the app → Open → Open anyway

Code signing is on the way. Until then the app's source is public, so you can see exactly what you're running: https://github.com/InferenceHub/inferencehub-desktop

Kick the tires and tell us what breaks — reply in the @inferencehub_group or open an issue. 🙏

Claude Opus 5 is live on InferenceHub: opus-5 on the API and Claude Code, and selectable in the hosted chat under Anthropic. Parity pricing, 0% markup - $5/$25 per 1M, $0.50 cached input, same sheet as Opus 4.8. Bare opus still means 4.8. Changelog: https://inferencehub.tech/changelog

Auto routing is live: name a price, not a model.

Two new model ids:
• ih-auto — routes each request to a fast open-weight model at a flat $1/M input, $5/M output. Never escalates.
• ih-auto-smart — same pool, but a genuinely hard turn escalates to Claude Sonnet 5, billed exactly as if you'd picked sonnet-5 yourself. Choosing -smart is the consent.

The honest print, up front:
• Named models here pass through at 0% markup. Auto is the one deliberate exception — the spread between the flat sheet and the routed model's native price is the routing fee.
• Not a black box: routing is a deterministic rule table (no LLM reads your prompt to classify it), and every response tells you what served it — the model field, the x-ih-auto header, and your usage view.
• On plans, Auto never touches the premium budget — the all-day coding id on Plus and Max.

Bonus for long agent sessions: the Auto ids can compact an overgrown conversation instead of erroring on context, and x-ih-compact: 1 trims to a working set on demand — you're billed on the compacted request.

Works everywhere a model id goes: Claude Code ("model": "ih-auto-smart"), Cursor (add as custom model), Kilo, plain API.

Docs: https://inferencehub.tech/docs/auto

📢 Big plans upgrade: one flat plan for chat, coding, and research

We've reorganized InferenceHub models into three clear tiers:

🔸 Premium (≥ $4/1M input tokens) — Fable 5, the Opus series, GPT 5.6 Sol
🔹 Standard ($1–$4/1M input tokens) — Sonnet 5, Kimi K3, GPT 5.6 Terra, GLM 5.2, and more
🆓 Free (< $1/1M input tokens, plus select frontier models) — Haiku 4.5, GPT 5.6 Luna, and more

Honest fine print: Free tier applies inside the chat app only. The same models used via the API (coding agents, integrations) are billed normally at list price.

⚡ Max plan — $50/month

- $500/month of premium model usage
- $2,000/month of standard model usage
- Works across everything: chat, deep research, and API/coding

For most users, that's your entire AI stack — chat + coding agent + research — covered by one $50 plan.

🧠 Prompt caching is now live across our models. Long conversations and coding sessions reuse cached context at a fraction of the cost, so your plan allowance stretches much further than raw token prices suggest.

Prefer something lighter? The Plus plan ($25/month) is also available — full breakdown: https://inferencehub.tech/pricing

Questions or edge cases? Feel free to ask in the group chat here: @inferencehub_group
We'll answer straight. Thanks all! 🙏

📣 InferenceHub is on Telegram — meet @InferenceHubBot. Link it to your account (portal → Account → Telegram) and chat with frontier + open-weight models on your own balance: same catalog, same pay-as-you-go pricing as the web chat and API. /models to browse, /model to switch, /new for a fresh thread. Free-class models included on trial. Nothing is sold in-chat — your portal stays in charge. https://inferencehub.tech/docs/telegram

🚀 Big open-weight drop on InferenceHub - 5 new models, all live:

🐋 DeepSeek V4 Flash (0731) - deepseek-v4-flash-0731
$0.08 in / $0.252 out per 1M · fast, agentic, absurdly cheap

🐋 DeepSeek V4 Pro (0813) - deepseek-v4-pro-0813
$1.32 in / $3.96 out per 1M · DeepSeek's flagship-class snapshot

🌀 Qwen3.8 Max - qwen3.8-max
$2.00 in / $6.00 out per 1M · Qwen's 2.4T-para

⚡ GLM-5.3 - glm-5.3
$1.40 in / $4.40 out per 1M · 1M context + up to 1M output (4× GLM-5.2)

⚡ GLM-5.3 Flash - glm-5.3-flash
$0.15 in / $0.50 out per 1M · same 1M window, fraction of the price

Every one of them is served natively on all three API surfaces - Chat Completions, Anthropic Messages, and Responses. That means Claude Code and Codex CLI work out of the box: swap the model id and go. Tool calling on all five, prompt caching where the model supports it.

One key. Every model. 🔑

Fable 5.1 is enabled now.

GPT-6 Astra is live on InferenceHub. OpenAI's new frontier model — 1.05M context, 128K output, native reasoning, tool calling, prompt caching — at pass-through pricing: $10 / 1M input, $50 / 1M output, $1 / 1M cached input.

Where it works today:
• API: gpt-6-astra on /v1/responses (Codex and any Responses client)
• Cursor: add gpt-6-astra as a custom model (Chat, Cmd-K, Agent)
• Claude Code: ANTHROPIC_MODEL=gpt-6-astra, or pick claude-code-gpt-6-astra from /model
• Chat: chat.inferencehub.tech → OpenAI → GPT-6 Astra

On plans it draws from the shared premium budget alongside Opus, Fable 5 and GPT-5.6 Sol.
Docs: https://inferencehub.tech/docs/models · Changelog: https://inferencehub.tech/changelog