Most free LLM API platforms stop at handing you an endpoint and a key. That's fine if you're wiring things into your own app, but what happens when you just want to quickly test a prompt against Qwen3, switch to GLM-5.2, and compare outputs side-by-side without writing a single line of code? That's where the UX gap lives — and that's exactly what apishare.cc just closed.
This week, Apishare shipped Agent Chat 2.0, a ground-up rebuild of its in-browser playground. It brings the kind of interaction polish you'd expect from a paid ChatGPT subscription — regenerate, edit-in-place, scroll-paginated history, vision uploads, a virtual file system, and even in-browser git clone — to a platform where every model is free at the point of use.
Let's walk through what shipped.
Why Free LLM Platforms Need Great UX, Not Just API Endpoints
Here's the uncomfortable truth about the current free LLM landscape: most platforms treat the playground as an afterthought. You get a text box, a send button, and maybe a model dropdown. Want to regenerate a response? Resend the entire conversation. Want to upload an image? Good luck. Want to reference a file? Copy-paste it in.
Developers don't just need API access — they need a fast iteration loop. You want to tweak a prompt, see the diff, try a different model, upload a screenshot, and do it all without context-switching between five tools. That's the bar ChatGPT set, and honestly, it's the bar every LLM platform should meet by now.
Agent Chat 2.0 is Apishare's answer to that gap.
Agent Chat 2.0 Features
Regenerate and Edit-in-Place
This is the feature that matters most for day-to-day use. Previously, regenerating a response meant resending the full conversation history — slow, wasteful, and disruptive if you're working with a long context window.
Now, regeneration works like ChatGPT's: click regenerate on any assistant message, and only that specific completion is re-requested. The conversation state stays intact. You can also edit a previous user message in-place, and the downstream responses update accordingly. No full resend, no lost context.
This alone changes how the playground feels. It goes from "API tester" to "actual thinking tool."
Scroll-Paginated History (40 messages/page, Viewport-Anchored)
Long conversations used to mean either slow initial load times or browser tab crashes. Agent Chat 2.0 implements scroll-based pagination at 40 messages per page, with viewport-anchored loading. As you scroll up, older messages load seamlessly without jumping the viewport.
Vision Image Upload (413 Fixed, Auto-Adapt)
Vision models need images, and image uploads are deceptively hard to get right. The previous version had a persistent 413 error when uploading larger images. That's fixed. Image uploads now auto-adapt: the client-side code detects image dimensions and file size, compresses when necessary, and sends in a properly chunked multipart payload.
Virtual File System (VFS)
Agent Chat 2.0 includes a Virtual File System that the agent can read from and write to during a conversation. Key capabilities:
- Read/write: The agent can read files you upload and write output files you can download
- grep search: The agent can search across files using grep-style pattern matching
- Line-numbered pagination: Large files are paginated with line numbers
This turns the playground into something closer to a lightweight IDE.
In-Browser Git Clone with Private Repo Support (PAT)
You can now clone a git repository directly in the browser. The clone lands in the VFS, and the agent can immediately start reading through the codebase. Private repositories are supported via Personal Access Tokens (PAT), which are stored per-session and never persisted server-side.
Per-User Storage Isolation
Every user gets an isolated storage namespace. Your files, your cloned repos, and your VFS state are completely separated from other users.
Trust Layer: Honest Billing + Withdrawal System
PayPal Double-Entry Confirmation + Balance Gates
Withdrawals now go through a double-entry confirmation system. When you request a withdrawal via PayPal, two independent entries are created and must reconcile before the transaction is processed. The balance gate checks your available balance before the request is queued and again before the payout is executed.
Register = Free Key, No Charge on Failure, GCRA Smooth Queueing
Every new registration gets a free API key immediately — no credit card, no trial period. If a request fails (provider timeout, model error, network issue), you are not charged. Billing only happens on successful completions.
Request queueing uses GCRA (Generic Cell Rate Algorithm) for smooth, fair scheduling.
Free LLM Leaderboard: 9 Providers Including SambaNova
Apishare now aggregates 9 providers on its public leaderboard: nvidia-nim, openrouter, groq, gemini, cloudflare, cohere, sambanova, siliconflow, zhipu.
Two providers that had degraded availability are now fully restored: Mistral and agnes-free are both back to 100% health.
The leaderboard is live at https://apishare.cc/leaderboard.
Available Free Models — This Week's Highlights
All models are free at the point of use.
ApiShare/alibaba/qwen3.8-max:free — Alibaba's flagship with 1M token context window. $0.
JamesFoster/tencent/Hunyuan-MT-7B — Multilingual translation specialist. $0.
MeiLin/THUDM/GLM-Z1-9B-0414 — Visible chain-of-thought reasoning. $0.
Agnes/agnes-3.0-flash:free — Lightning-fast response model. $0.
ApiShare/zhipu/glm-5.2:free — Advanced reasoning. $0.
NEW: SiliconFlow GLM-4-9B and PaddleOCR-VL-1.5 — Just added this week.
Getting Started: Quick Code Example
curl -X POST https://apishare.cc/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $APISHARE_API_KEY" \
-d '{
"model": "ApiShare/alibaba/qwen3.8-max:free",
"messages": [
{
"role": "user",
"content": "Explain the difference between GCRA and token bucket rate limiting."
}
],
"max_tokens": 500
}'
The API is OpenAI-compatible — any existing client library that speaks the OpenAI Chat Completions format will work. Point the base URL at https://apishare.cc/v1 and use your Apishare API key.
Agent Chat 2.0 is live right now. Registration takes about 10 seconds and you get a free API key immediately.
Try it free at apishare.cc.
What would you build?
Top comments (0)