FREE GPT-6 Astra (Sep 2026): $200 OG credits, no card — one API key for GPT-6 Astra (1.05M ctx), GPT-5.5, Claude Opus 4.8, GLM 5.2, DeepSeek R1, Gemini 2.0 Pro + 30 models. OpenAI-compatible. New model
gpt-6-astralive — 1.9x faster computer use, 99.9% ARC-AGI-3. Try FREE via my referral before pool resets.
Register via my link to get OG ($200) credits instead of $100 — FREE access to GPT-6 Astra + GPT-5.5 + Claude:
https://agentrouter.org/register?aff=v3aGStandard = $100 | Referral = $200 OG — unlocks FREE GPT-6 Astra (1.05M ctx), GPT-5.5, Claude Opus 4.8, GLM, DeepSeek, Gemini. I get a referral bonus to keep this guide free. Hype drop: GPT-6 Astra is live (Sep 3, 2026) — world's most intelligent, 99.9% ARC-AGI-3, 100% ExploitBench. Try it FREE via AgentRouter, one key, no card. Pool resets early (SG time), claim now. Seedocs/10-gpt6-astra.md.
Why this is going viral — click to expand
- FREE new GPT model without OpenAI Pro/Max — via AgentRouter gateway
- FREE Claude Opus 4.8 (1M context) without Anthropic Max — same key
- FREE GLM 4.5 Air, DeepSeek V2 Lite — $0 forever for routing/autocomplete
- No credit card, GitHub OAuth only — credits land instantly
- One key, every AI — Codex, OpenCode, Claude Code, Cursor, Cline, Roo, Aider, LangChain all work
Built for the Astra hype cycle: new GPT drops -> everyone wants to try -> AgentRouter lets you try for $0.
This repo is an independent, community-maintained guide. Not affiliated with AgentRouter. Verified against live docs on 2026-09-12.
Viral kit: See VIRAL.md for tweet / Reddit / HN / Discord copy-paste templates to hype the free new GPT launch.
AgentRouter (agentrouter.org) is an OpenAI-compatible unified LLM gateway launched Oct 2025 (China, Singapore infra). It proxies requests to upstream providers behind a single endpoint — you change only baseURL + apiKey, not your code.
Your App ──(OpenAI JSON over HTTPS)──▶ AgentRouter Gateway ──┬──▶ Anthropic (Claude)
│ single key / single bill • auth & routing ├──▶ OpenAI (GPT-5.5)
└─────────────────────────────────────• accounting ├──▶ Zhipu (GLM 5.2)
• queuing ├──▶ DeepSeek / Gemini / Qwen ...
└──▶ +30 more
Core promises:
- Unified Interface — OpenAI
/v1/chat/completions+ Anthropic/v1/messagescompatible. Swap models by changing one string. - Intelligent Routing — auto load-balancing, failover, multi-provider redundancy (99.9% claimed).
- Cost Optimization — pay-per-token, no subscription, ~27% of public list price claimed, volume discounts, pay-as-you-go, balance never expires.
- Enterprise Portal —
co.agentrouter.orgprovides admin console, SSO/RBAC, quotas, audit logs, data-governance controls (fine-grained provider routing).
One-line pitch from the homepage: “Better price, better stability, no subscription required, just replace the model BASE URL.” —
agentrouter.org· Built on New API & One API.
| Domain | What it actually is | Base URL | Relation to this guide |
|---|---|---|---|
agentrouter.org |
LLM gateway (THIS GUIDE) — New API fork | https://agentrouter.org/v1 |
Primary subject |
co.agentrouter.org |
Enterprise portal for same gateway (pricing, models, rankings, portal/guide) |
https://co.agentrouter.org (Anthropic) / .../v1 (OpenAI) |
Same infra, richer docs — treated as canonical |
agent-router.org |
MCP marketplace for specialist agents (Tim Brauer) — One MCP URL → 800+ agents |
mcp.agent-router.org |
Different product, unrelated |
agentrouter.to / docs.agentrouter.to |
Capability-first gateway for paid APIs (email, search, web, etc.) with recommend/execute + wallet |
https://api.agentrouter.to |
Different product, see intro.quickstart there |
If you landed here looking for MCP agent delegation or domains/capabilities/wallet APIs — you want the other two domains. This guide covers the LLM gateway only.
- My referral (use this):
https://agentrouter.org/register?aff=v3aG(GitHub OAuth, no credit card)- Gives you $200 OG credits (vs $100 standard) — I get a small referral bonus
aff=v3aGmust be in URL at registration — can't be added later
- Copy key at
https://agentrouter.org/console/token→sk-... - Support the guide: share
https://agentrouter.org/register?aff=v3aG— you get $200, I get +$100 to keep this guide updated
curl https://agentrouter.org/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-YOUR_KEY" \
-d '{
"model": "claude-sonnet-4-5-20250929",
"messages": [{"role": "user", "content": "Say hello in one sentence"}],
"max_tokens": 50
}'from openai import OpenAI
client = OpenAI(api_key="sk-YOUR_KEY", base_url="https://agentrouter.org/v1")
resp = client.chat.completions.create(
model="claude-sonnet-4-5-20250929",
messages=[{"role": "user", "content": "Write a haiku about APIs"}]
)
print(resp.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({ apiKey: process.env.AGENTROUTER_API_KEY!, baseURL: "https://agentrouter.org/v1" });
const r = await client.chat.completions.create({ model: "gpt-5.5", messages: [{role:"user", content:"ping"}] });
console.log(r.choices[0].message.content);Anthropic-compatible (Claude Code, etc.): base URL is https://co.agentrouter.org or https://agentrouter.org without /v1.
OpenAI-compatible (everything else): base URL must end with /v1.
| Doc | What’s inside |
|---|---|
| GUIDE.md | Single-file definitive guide (recommended read) |
docs/01-introduction.md |
Philosophy, who should/shouldn’t use it |
docs/02-architecture.md |
How the thin-proxy gateway works, latency, reliability |
docs/03-models-pricing.md |
Catalog, pricing tiers, credit burn modeling |
docs/04-quickstart.md |
Registration → key → first call, step-by-step |
docs/05-api-reference.md |
Endpoints, auth, streaming, errors |
docs/06-integrations.md |
Every client: Claude Code, Codex, OpenCode, Cursor, Cline, Roo, Kilo, Copilot, Trae, Craft… |
docs/07-examples.md |
Code review bot, benchmark harness, fallback pattern |
docs/08-troubleshooting.md |
401/404/402, quota exhausted, GitHub auth issues |
docs/09-providers.md |
All providers (Anthropic/OpenAI/GLM/DeepSeek/Gemini/Qwen/Mistral) for Codex, OpenCode, every AI |
docs/10-gpt6-astra.md |
NEW: GPT-6 Astra (Sep 2026) — FREE via AgentRouter, 1.05M ctx, benchmarks, Codex/OpenCode configs |
VIRAL.md |
Viral kit — tweet / Reddit / HN / Discord templates for GPT-6 Astra hype |
Portal highlights (co.agentrouter.org/portal/models):
| Model | Provider | Context | Input / 1M | Output / 1M | Best for |
|---|---|---|---|---|---|
| GPT-6 Astra | OpenAI | 1.05M (922K in / 128K out) | $10 | $50 | NEW Sep 2026 — best for computer use, coding, science |
| Claude Opus 4.8 | Anthropic | 1M | $8 | $40 | Deep agent, long-horizon tasks |
| Claude Opus 4.7 | Anthropic | 1M | $8 | $40 | Async agents, large codebase |
| Claude Opus 4.6 | Anthropic | 1M | $2 | $10 | Stable enterprise default |
| GPT-5.5 | OpenAI | 1M (922K in / 128K out) | $4 | $8 | Reasoning, coding, multimodal |
| GLM 5.2 | Zhipu | 1M | $3 | $4.5 | Cost-effective engineering |
- via gateway:
gpt-6-astra(NEW 99.9% ARC-AGI-3, 100% ExploitBench),claude-sonnet-4-5-20250929,claude-haiku-4-5-*,deepseek-r1,gemini-2.0-pro,glm-4.5-air(free),qwen3-coder-480b, etc. Full list →docs/03-models-pricing.mdand newdocs/10-gpt6-astra.md.
- Free credits (OG via my referral):
https://agentrouter.org/register?aff=v3aG→ $200 OG (vs $100 standard) — no expiry quoted, GitHub OAuth only (accounts <1 yr may fail). You get $200, I get +$100 referral bonus. - Pay-as-you-go: no weekly caps, no minimum, balance never expires
- Enterprise: custom quote (
neo@agentrouter.org), volume discounts, multi-tenant admin, SSO/RBAC, audit logs - Claimed edge: ~27% of public API list price for equivalent usage (portal)
Heads-up from community reports: even with credits remaining you may hit
402 Budget pool quota has been exhausted— a per-model global rate limit. Fix: switch model or retry later. See troubleshooting.
From agentrouter.org homepage:
/v1/chat/completions /v1/responses /v1/messages /v1beta/models
/v1/embeddings /v1/rerank /v1/images/generations /v1/images/edits
/v1/images/variations /v1/audio/speech /v1/audio/transcriptions /v1/audio/translations
All are OpenAI-shape passthroughs.
VS Code: Claude Code for VS Code, Cline, Roo Code, Kilo Code, GitHub Copilot CLI/TUI: Claude Code CLI, Codex, opencode, Qwen Code, Crush, Hermes Agent Desktop: Claude App (Gateway), Trae, Cursor, Craft Agents Frameworks: LangChain/LangGraph, LlamaIndex, Continue.dev, n8n
One key works across all — quota is shared. See docs/06-integrations.md for copy-paste configs.
- Production systems needing an SLA / incident response
- HIPAA/GDPR/regulated data (no DPA publié, proxy sees prompts)
- Ultra-low latency (<50ms) — Singapore hop adds 80–150ms to US/EU
- You need a web chat UI — gateway is API-only
Anthropic vs OpenAI base URL?
Anthropic (Claude): https://co.agentrouter.org (no /v1). OpenAI (GPT etc.): https://co.agentrouter.org/v1 or https://agentrouter.org/v1 — don’t mix.
One key for multiple agents?
Yes. Cline + Roo + Claude Code can share the same sk-...; usage is unified.
Do I modify source code?
No. Only base_url/baseURL + api_key.
GitHub auth fails?
New GitHub accounts (<1 yr) are often rejected. Use older account or try gorouter.app mirrors.
PRs welcome. Keep edits factual and cite the portal or live test. Run npm run lint if you add code samples.
MIT — this guide is community content. AgentRouter branding belongs to its owners.
- My OG referral: https://agentrouter.org/register?aff=v3aG ← use this for $200 OG credits (supports this guide)
- Gateway: https://agentrouter.org · Console: https://agentrouter.org/console/token
- Enterprise portal: https://co.agentrouter.org · Docs: https://co.agentrouter.org/portal/guide
- Contact:
neo@agentrouter.org· QQ Group ·x.com/AgentRouter_0· Discorddiscord.gg/WcwGwcHaq
Built with research on 2026-09-12. Prices/models change — verify at
/portal/modelsand/portal/pricingbefore billing.