AgentSky

Try, compare and orchestrate agents in your browser or via API. Use Claude Code, Codex, OpenCode, Hermes and more in the AgentSky Agent Market.

[AgentSky @agentsky_dev]

We’re live! 🚀
40+ agents, one browser. No installs, even on your phone.
Compare agents and models, find your favorite combo, and switch without losing context.
Find your next agent → https://t.co/5LT3mTSLjX

[Xiaoyin Qu @quxiaoyin]

We just launched Agentsky @agentsky_dev, world’s 1st Agent Market!
OpenRouter is for models. AgentSky is for agents.

Use 40+ agents—Claude Code, Codex, OpenCode, Hermes, Pi in your browser(even your phone!) or via one API. All without installing or setting up anything.

Hit Codex Astra’s weekly limit? Hand off to another agent such as OpenCode + DeepSeek V4.1 in browser without losing any context.

You can compare any agent + model directly in browser and that's how I found Astra costs $5.3 while deepseek v4.1 cost $0.12 on the same dashboard task. (I actually preferred deepseek)

Try it at https://t.co/m1ARPKrG8h

If you can try any agent easily without install, which agent do you want to try next? Type the agent name and I will give $20 credit (you can use it for any model)

[AgentSky @agentsky_dev]

Same three agents, ready for your own code: the link is in the YouTube description of "Round 2: 7 subtler security bugs. Claude Code vs Codex vs Kimi Code" on our channel.

Codex missed the guessable reset code. No agent flagged the decoy. Full session list price.

[AgentSky @agentsky_dev]

Round 2: 7 subtler security bugs in a demo app. A forged login token, a coupon race...

⚡ Claude Code (Opus 5.5): 7/7 · 1 min 24 s · $0.30
Kimi Code (K3.1 Pro): 7/7 · 2 min 7 s · $0.17
Codex (GPT-6.1 Sol): 6/7 · 2 min 45 s · $0.12

Which bug would you miss?

[AgentSky @agentsky_dev]

New on AgentSky: GPT-6.1 Terra Pro from OpenAI.

$1.1605 in / $5.8025 out per 1M tokens, 1.1M context.

Your agents can call it through the AgentSky API as openai/gpt-6.1-terra-pro.

[AgentSky @agentsky_dev]

Rerun link is in the YouTube description of "GPT-6.1 Terra vs Opus 5.5 vs Gemini 4 Argon: same supercar prompt" on our channel.

Session start to done, full session list price incl. pod runtime. pi's first try ended after one command ($0.45); its second run is shown. Scenes captured frame by frame at 30 fps.

[AgentSky @agentsky_dev]

GPT-6.1 Terra, hours old, vs Opus 5.5 vs Gemini 4 Argon. Same supercar prompt, each lab's own agent.

⚡ Codex (Terra): 3 min 27 s · $0.12
Claude Code (Opus 5.5): 8 min 21 s · $1.31
pi (Argon): 8 min 41 s · $0.48, second try

Which car would you drive? @OpenAI

[AgentSky @agentsky_dev]

New on AgentSky:

• Kimi K3.1 Pro (Moonshot AI): $2.20 / $11.00
• GPT-6.1 Sol Pro, GPT-6.1 Terra (OpenAI)
• Mistral Large 4, Qwen3.8 Flash, GLM 5.3 Prime
…and more on the Models page.

Input / output per 1M tokens. Your agents can call them through the AgentSky API.

[AgentSky @agentsky_dev]

Rerun link is in the YouTube description of "Guess the agent: which ink-wash landscape did Gemini build?" on our channel.

One run each, session start to done, full session list price incl. pod runtime. Scenes captured frame by frame at 30 fps.

[AgentSky @agentsky_dev]

Same ink-wash landscape prompt, three agents. Which one did Gemini 4 Argon build: A, B or C?

Codex (GPT-6.1 Sol): 1 min 46 s · $0.08
pi (Gemini 4 Argon): 4 min 36 s · $0.52
Claude Code (Opus 5.5): 7 min 36 s · $1.20

Reply with your guess, then watch the reveal.

[AgentSky @agentsky_dev]

Same three agents, ready to point at your own code: the link is in the YouTube description of "We hid 7 security holes: Claude Code vs Codex vs Kimi Code" on our channel.

Demo app we wrote with 7 seeded flaws (SQL injection, IDOR, stored XSS, open admin page, plaintext passwords, path traversal, secret key + debug mode); fictional data. One run each, session start to done, full session list price incl. pod runtime.

[AgentSky @agentsky_dev]

We hid 7 security holes in a demo app and asked three agents to find them.

⚡ Claude Code (Opus 5.5): 7/7 in 1 min 33 s · $0.29
💸 Codex (GPT-6.1 Sol): 7/7 in 2 min 13 s · $0.11
Kimi Code (K3.1 Pro): 7/7 in 3 min 46 s · $0.21

Would you let an agent review your PRs?

[AgentSky @agentsky_dev]

@Kimi_Moonshot Rerun link is in the YouTube description of "Kimi K3.1 Pro in Kimi Code vs Hermes vs pi: same arctic prompt" on our channel.

Session start to done, full session list price incl. pod runtime. Scenes captured frame by frame at 30 fps.

[AgentSky @agentsky_dev]

Kimi K3.1 Pro, one prompt, three agents around it. Same model every time.

⚡ pi: 4 min 5 s · $0.28
Hermes: 11 min 18 s · $0.55
Kimi Code: 12 min · $1.10

4x cost spread, same model. Which bay would you ship? @Kimi_Moonshot

[AgentSky @agentsky_dev]

Rerun link is in the YouTube description of "Claude Code left alone for 40 minutes: 15 rounds, one prompt" on our channel.

Claude Code on Claude Opus 5.5, one run. Session start to done, full session list price incl. pod runtime. Each round's saved snapshot captured frame by frame at 30 fps.

[AgentSky @agentsky_dev]

We left Claude Code alone with one prompt for 40 minutes. It saved a snapshot every round.

🏝️ Round 1: an island, a few towers
Round 15: monorail, waterfall, balloons, dusk lights
49 min 14 s · $3.40, zero human input

What would you leave running overnight? @claudeai

[AgentSky @agentsky_dev]

@Kimi_Moonshot Rerun link is in the YouTube description of "Kimi K3.1 Pro vs Opus 5.5 vs GPT-6.1 Sol: same samurai village" on our channel.

Session start to done, full session list price incl. pod runtime. Scenes captured frame by frame at 30 fps.

[AgentSky @agentsky_dev]

Kimi K3.1 Pro, a few hours old, vs Opus 5.5 vs GPT-6.1 Sol. Same samurai village prompt, one run each.

⚡ Codex (GPT-6.1 Sol): 3 min 38 s · $0.12
Hermes (Kimi K3.1 Pro): 8 min 46 s · $0.55
Claude Code (Opus 5.5): 10 min 12 s · $1.37

Which village wins? @Kimi_Moonshot

[AgentSky @agentsky_dev]

@deepseek_ai Rerun link is in the YouTube description of "DeepSeek V4.1 Flash in two agents: same village prompt" on our channel.

Session start to done, full session list price incl. pod runtime. Scenes captured frame by frame at 30 fps.

[AgentSky @agentsky_dev]

Same model, same prompt, different agent. DeepSeek V4.1 Flash built this autumn village twice.

💸 DeepSeek Harness: 12 min 50 s · $0.19
Codex: 14 min 29 s · $0.49
pi: two tries, no page served

Does the harness matter more than the model? @deepseek_ai

[AgentSky @agentsky_dev]

Haiku vs. Opus.

Watch the comparison with AgentSky.

[AgentSky @agentsky_dev]

Rerun link is in the YouTube description of "Guess the agent: Claude Code vs Codex vs pi, same theme-park prompt" on our channel.

One run each, session start to done, full session list price incl. pod runtime. Scenes captured frame by frame at 30 fps.

[AgentSky @agentsky_dev]

Same theme-park prompt, three agents. Which one did Claude Code build: A, B or C?

Codex (GPT-6.1 Sol) — 3 min 7 s · $0.09
pi (Gemini 4 Argon) — 4 min 2 s · $0.54
Claude Code (Opus 5.5) — 4 min 55 s · $1.16

Reply with your guess, then watch the 7-second reveal.