When we first published this Kimi AI review, the headline wrote itself: a hundred free AI agents against ChatGPT’s $200 subscription. That story is over. Kimi K2.6 is a stronger model than the K2.5 we tested, its Agent Swarm now coordinates up to 300 sub-agents, and Moonshot AI is reportedly worth $20 billion. But the free ride quietly ended: agent features moved behind a subscription ladder that runs from $19 to $199 a month, and the API costs several times what it did in winter. This updated Kimi AI review covers what K2.6 actually does, what every tier costs, and whether the paywalled version still beats ChatGPT, DeepSeek, and Gemini.
⚡ TL;DR – The Bottom Line
What It Is: Moonshot AI’s open-weight multimodal flagship (K2.6) with chat, coding, and a 300-agent swarm mode, delivered through kimi.com and an API.
Best For: Power users who want frontier-class agents cheaper than ChatGPT Pro, and developers who want open weights with a hosted API.
Price: Free basic chat (Adagio); paid tiers $19 / $39 / $99 / $199 per month; API $0.95 per million input tokens, $4.00 output.
Our Take: Our Kimi AI review verdict: still the best capability-per-dollar in AI agents, but the “everything free” era is finished. Budget for at least the $19 tier if you came for agents.
⚠️ The Catch: The free tier now allows roughly 6 agent tasks and excludes Agent Swarm and Kimi Code entirely. Benchmarks below are Moonshot’s own numbers.
📑 Quick Navigation
🚀 What Changed: K2.5 → K2.6 (and the End of Free)
Two shifts define this update. First, the model: K2.6 arrived in April 2026 as the new open-weight flagship, keeping native multimodality and pushing Moonshot’s reported scores higher across the board (Humanity’s Last Exam with tools went from 50.2 to 54.0). Second, the business model: the anything-goes free access that made Kimi famous is gone. A five-tier subscription ladder now gates the agent features, and the API moved to a new home at platform.kimi.ai with substantially higher prices. Moonshot also reportedly raised $2 billion at a $20 billion valuation in May, which explains both the pace and the paywalls.
🧠 What Kimi Actually Does
Kimi is three products wearing one interface. It is a chat assistant that sees images and video, in the same category as ChatGPT and Gemini. It is an agent platform, where tasks run unattended in the cloud: Agent mode for single jobs, Agent Swarm for parallel armies, plus the newer Kimi Claw for always-on cloud agents and Kimi Work, a desktop agent. And it is an open-weight model family developers can self-host or hit through the API, which is why any serious Kimi AI review has to grade it against both consumer chatbots and agent frameworks at once.
📱 Getting Started on the Free Tier
Every Kimi AI review should start where users do: signup. At kimi.com it remains painless: email or Google login, no phone number gymnastics, and the mobile apps mirror the web experience. What changed is what waits behind the door. The free Adagio tier still gives you unlimited basic chat with the flagship model, which remains a remarkable deal. But agent work is now a taste test: roughly 6 agent-task uses, one concurrent task, and no access to Agent Swarm or Kimi Code at all. Treat free as an evaluation lane, not the product.
🐝 Agent Swarm: Now 300 Workers at Your Command
The signature feature tripled. Agent Swarm on K2.6 coordinates up to 300 sub-agents across as many as 4,000 steps, up from 100 agents and 1,500 steps in K2.5. In practice that means one prompt can fan out into parallel researchers, coders, and reviewers that reconcile their work into a single deliverable. Moonshot’s own numbers show why they lead with it: BrowseComp rises from 83.2 to 86.3 when the swarm is on. The catch sits in the billing table: swarm access starts at the paid tiers and is metered in monthly “uses” (from 25 on the $19 plan to 240 on the $199 plan), so the hundred-agent free-for-all we described in winter is now a budgeted resource.
💡 Key Takeaway: Count your real swarm needs before picking a tier. Most solo users burn 5 to 15 swarm uses a month, which the $19 Moderato tier covers. The $199 tier only makes sense if agents are your daily workflow, not your weekend experiment.
💭 Thinking Mode and the Benchmark Story
K2.6 ships thinking and non-thinking variants in one model, with a 262,144-token context window. Moonshot’s reported results put it at or near the top of the open-weight class: 54.0 on Humanity’s Last Exam with tools, 80.2 on SWE-Bench Verified, 90.5 on GPQA-Diamond, and 96.4 on AIME 2026. Every one of those is a vendor-published number, so apply the usual discount, but the direction matches what independent users report: this is a frontier-adjacent model with open weights on HuggingFace.
🎯 K2.6 Vendor-Reported Benchmarks (July 2026)
📊 Slides, Sheets, and the New Kimi Work
The office layer keeps growing: Slides, Websites, and Sheets generation live alongside the agents, and the new Kimi Work desktop agent points at the same automation territory as Microsoft’s Copilot. The slides generator remains the most polished of the set. We keep our deep dive separate in the Kimi AI Slides review; the short version is that it still produces the best AI-native deck drafts we have tested at this price, and it now draws from the same metered agent quota as everything else.
💰 Pricing Breakdown: The Free Ride Ends
The pricing section of this Kimi AI review changed more than any other. Verified July 2026 from the official membership page. The tier names are musical tempo markings, which is charming and slightly confusing:
| Tier | Monthly | Effective Monthly (Annual) | What It Gates |
|---|---|---|---|
| Adagio (Free) | $0 | $0 | Unlimited basic chat; ~6 agent tasks; no Swarm, no Kimi Code |
| Moderato | $19 | $15 | Agent Swarm entry quota (~25 uses), Kimi Code access |
| Allegretto | $39 | $31 | Higher agent + swarm quotas, more concurrency |
| Allegro | $99 | $79 | Heavy agent quotas for daily professional use |
| Vivace | $199 | $159 | Top quotas (~240 swarm uses), maximum concurrency |
The real-world cost math:
- Per swarm use on Moderato: $19 ÷ ~25 uses = ~$0.76 per swarm run
- Per swarm use on Vivace: $199 ÷ ~240 uses = ~$0.83 per swarm run (you pay for concurrency and quota height, not a better rate)
- API, K2.6: $0.95 per million input tokens (cache miss), $0.16 cache hit, $4.00 per million output, 262K context
- Kimi Vivace ($199) vs ChatGPT Pro ($200): near price parity at the top, a sentence nobody would have written in January
📊 Kimi Subscription Ladder (Monthly vs Annual-Effective)
For developers, the API move stings more than the consumer tiers. The K2.6 rate of $0.95 in and $4.00 out is several times winter’s K2.5 pricing, and the platform migrated from platform.moonshot.ai to platform.kimi.ai in the process. It is still cheap next to Western frontier APIs, and the open weights mean self-hosting remains the ultimate price cap, but the era of rounding-error API bills is over.
🔍 REALITY CHECK
Marketing Claims: “100 free AI agents” was the pitch that made Kimi famous (and our old headline).
Actual Experience: The free tier now includes roughly 6 agent tasks, one at a time, with Agent Swarm and Kimi Code excluded entirely. The swarm exists, is bigger than ever at 300 sub-agents, and starts at $19/month with metered uses.
Verdict: Judge Kimi as a $19-to-$199 product with a generous free chat tier, not as a free agent platform. It still wins that comparison, but it is a different comparison.
📬 Enjoying this review?
Get honest AI tool analysis delivered weekly. No hype, no spam.
⚔️ Kimi vs the Giants (July 2026 Field)
No Kimi AI review means much without the field around it, and the field moved as fast as Kimi did: OpenAI shipped GPT-5.5, Anthropic is on Claude Opus 4.7, DeepSeek landed V4, and Google runs Gemini 3.1 Pro. Where Kimi still stands alone is the combination: open weights, a consumer agent platform, and pricing that undercuts the West at every tier.
| Criteria | Kimi K2.6 | GPT-5.5 / ChatGPT | DeepSeek V4 | Winner |
|---|---|---|---|---|
| Open weights | Yes (HuggingFace) | No | Yes | Kimi / DeepSeek |
| Consumer agent platform | Swarm, Claw, Kimi Work | Agent mode, Codex | Minimal | Kimi |
| Top-tier price | $199 (Vivace) | $200 (Pro) | Free/API-first | DeepSeek on price |
| Ecosystem & polish | Improving fast | Deepest ecosystem | Sparse | ChatGPT |
| Data residency comfort (Western orgs) | Chinese company | US | Chinese company | ChatGPT |
For coding specifically, the calculus involves tools rather than just models: Claude Code and the developer agent field compete with Kimi Code on workflow, not benchmark points. And if your interest is pure research assistance, Perplexity remains the simpler answer.
🔒 Privacy and Data Residency
Nothing structural changed here, and the consideration is the same one we flagged in every prior version of this review: Moonshot AI is a Beijing-based company, and consumer chats flow through its infrastructure. For personal projects and public research, most users accept that trade for the price. For client work, regulated data, or anything covered by a confidentiality agreement, use the open weights on your own hardware or skip Kimi for that workload. The open-source release is the honest escape hatch here, and it is one ChatGPT does not offer.
❓ Kimi AI Review FAQs: Your Questions Answered
Q: Is Kimi AI still free?
A: Basic chat, yes, and it is still generous. Agents, no: the free Adagio tier allows roughly 6 agent tasks and excludes Agent Swarm and Kimi Code. Paid tiers run $19 to $199 per month, about 20% less on annual billing.
Q: What happened to the 100 free AI agents?
A: The swarm grew to 300 sub-agents but moved behind the subscription ladder. Swarm uses are metered monthly, roughly 25 on the $19 tier up to 240 on the $199 tier.
Q: What is Kimi K2.6?
A: Moonshot’s flagship since April 2026: open-weight, natively multimodal, thinking and fast modes in one model, 262K-token context. Moonshot reports 54.0 on Humanity’s Last Exam with tools and 80.2 on SWE-Bench Verified. A coding-focused K2.7 Code variant followed in June.
Q: How much does the Kimi API cost?
A: For K2.6 on platform.kimi.ai: $0.95 per million input tokens on a cache miss, $0.16 on a cache hit, and $4.00 per million output tokens. That is a real increase over winter’s rates, though still below Western frontier APIs.
Q: Can Kimi replace ChatGPT?
A: For chat, research, and agent experiments, yes, and usually cheaper. ChatGPT keeps the edge on ecosystem depth, third-party integrations, and data-residency comfort for Western organizations.
Q: Is my data safe with Kimi?
A: Moonshot is a Beijing-based company and hosted chats run on its infrastructure. Keep confidential or regulated work off the hosted product; the open weights let you self-host when data control matters.
Q: Kimi or DeepSeek?
A: Both are open-weight Chinese labs. DeepSeek V4 is the raw-value API play; Kimi is the one with a consumer agent product around the model. If you want an app, Kimi; if you want cheap tokens, compare API sheets.
Q: What is the learning curve like?
A: Chat is instant. Agent and Swarm modes take a few sessions to prompt well, mostly learning to scope tasks tightly so metered uses are not wasted on vague requests.
🏆 Final Verdict
Our updated Kimi AI review lands on the same score for opposite reasons. In winter, Kimi earned 4.5 stars by giving away frontier agents for free. In summer, it earns 4.5 stars because K2.6 is a legitimately stronger product whose $19 entry tier still embarrasses the per-seat math of every Western rival. The half-star deduction moved: it used to be polish and uncertainty; now it is the quiet speed of the paywall’s arrival and vendor-only benchmarks.
✅ What We Liked
- ✓ K2.6: frontier-adjacent, open weights, 262K context
- ✓ Agent Swarm at 300 sub-agents is unmatched at the price
- ✓ $19 entry tier undercuts every Western agent product
- ✓ Free chat tier remains genuinely useful
- ✓ Open weights = self-hosting escape hatch for privacy
❌ What Fell Short
- ✗ The free agent era ended with little notice
- ✗ API prices multiplied (now $0.95/$4.00 per M tokens)
- ✗ Benchmarks are vendor-reported
- ✗ Beijing data residency rules it out for regulated work
Use Kimi if: you want the strongest agent platform under $20/month, or open weights you can actually run. Stick with ChatGPT if: ecosystem, integrations, and Western data residency outweigh price. Try the free tier at kimi.com, and check the current quotas on the official membership page before subscribing, because this pricing has changed twice in six months.
Founder of AI Tool Analysis. Tests every tool personally so you don’t have to. Covering AI tools for 10,000+ professionals since 2025. See how we test →
Related Reading
- Kimi AI Slides Review
- DeepSeek vs ChatGPT: The Open-Weight Battle
- ChatGPT Review
- Gemini 3 Review
- Grok 4 Review
- AI Agent Frameworks Compared
- Claude Code Review
- The Complete AI Tools Guide
Stay Ahead of the AI Curve
Don’t miss the next big update. Get AI Weekly delivered every Thursday at 9 AM EST with honest reviews, price drops, and features you can actually use.
- ✅ Honest Reviews: We actually test these tools, not rewrite press releases
- ✅ Price Tracking: Know when tools drop prices or add free tiers
- ✅ Feature Launches: Major updates covered within days
- ✅ Comparison Updates: As the market shifts, we update our verdicts
- ✅ No Hype: Just the AI news that actually matters for your work
Free, unsubscribe anytime. 10,000+ professionals trust us.
Last Updated: July 3, 2026
Kimi Version Tested: Kimi K2.6 (web, mobile, and API)
Next Review Update: October 2026 (or on K3 release)
Have a tool you want us to review? Suggest it here | Questions? Contact us