RXed AI News

Your AI intelligence briefing
SCORE 5.6/1016/20 elements Audited 2026-07-13 · RXed table v1.0

Manus

The most capable general-purpose autonomous agent on the market — genuine multi-step task execution in a real sandbox VM — but opaque credit economics, unresolved Meta-acquisition turmoil and thin compliance make it a bet, not infrastructure.

SpecialistAutomation & AgentsResearchCodingFreemium
Vendor
Butterfly Effect Pte. Ltd. (now part of Meta, acquisition contested by Chinese regulators) · manus.im
Origin
Singapore — Singapore
Pricing
Free (300 daily refresh credits) · Standard ~€19/mo (4,000 credits) · Pro ~€37/mo (8,000 credits) · Extended ~€185/mo (40,000 credits + free cloud computer); Team plans custom; annually 17% off
Users (official only)
Millions of active users; 2M+ waitlist within 7 days of launch; $100M ARR 8 months post-launch (source, 2025-12-17)

Element scores

Reactive
Retrieval & Memory
Orchestration
Validation
Models
Primitives
Pr6
Prompts
Em
Embeddings
Cx7
Context
Tr6
Tracing
Lg7
LLM
Compositions
Fc7
Function calling
Vx
Vector store
Rg6
RAG
Gr4
Guardrails
Mm6
Multimodal
Deployment
Ag8
Agents
Ft
Fine-tuning
Fw5
Frameworks & harnesses
Ev3
Evaluations
Sm5
Small models
Emerging
Ma7
Multi-agent
Sy
Synthetic data
Pc4
Protocols
In3
Interpretability
Th6
Thinking models
Tap or hover any element to see why it got that score.

Strengths

The most capable general-purpose autonomous agent available: genuine multi-step task execution in a sandboxed cloud VM with real browser, real terminal and persistent file system. Multi-agent architecture decomposes complex goals into subgoals with specialist agents. Wide Research runs parallel analysis across dozens of sources. Asynchronous execution means you can close the tab and come back to finished work. Manus Desktop (Mar 2026) bridges cloud and local machine. $100M ARR in 8 months — fastest startup to that milestone. Public REST API, Skills system, 100+ Playbook templates, Zapier integration, Slack and email integrations. Topped the GAIA benchmark (86.5% on Level 1, beating OpenAI Deep Research).

Honest dings

Credit economics are opaque — users routinely report surprise bills and a single complex task can consume thousands of credits. Reliability degrades sharply on long or branching tasks: loops, repeated steps and incomplete deliverables are widely documented. The Meta acquisition (announced Dec 2025, ~$2B) was blocked by Chinese regulators in Apr 2026, creating unresolved ownership and continuity risk — the site still says '© 2026 Meta'. No SOC 2 or GDPR at launch; the pricing page now claims SOC 2 Type II and ISO 27001 but the trust center timed out and compliance remains thin compared to enterprise alternatives. Server instability during peak hours. Coding quality trails dedicated coding agents (Cursor, Claude Code). No team collaboration on lower tiers.

Best for: Researchers, founders and operators who want an autonomous AI agent for multi-step web research, rapid prototyping and recurring admin tasks — and can absorb unpredictable credit costs and the Meta-ownership uncertainty.

Sources

Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).

  • manus.im — Official homepage: autonomous general AI agent, sandbox VM, now part of Meta, product features list (accessed 2026-07-13)
  • manus.im/pricing — Official pricing page (browser-verified): Free, Standard $20/mo (4,000 credits), Pro $40/mo (8,000 credits), Extended $200/mo (40,000 credits + free cloud computer); 300 daily refresh credits; annual 17% off; SOC 2 Type II + ISO 27001 claimed (accessed 2026-07-13)
  • manus.im/blog/manus-100m-arr — Official blog: $100M ARR 8 months post-launch, $125M revenue run-rate, 147T tokens processed, 80M+ virtual computers created, 105 employees across Singapore/Tokyo/SF, Benchmark-led $75M round (accessed 2026-07-13)
  • en.wikipedia.org/wiki/Manus_(AI_agent) — Wikipedia: founded by Butterfly Effect (China/Singapore), launched Mar 6 2025, Manus 1.6 stable Dec 2025, Meta acquisition Dec 2025 ~$2B, Chinese regulators ordered unwind Apr 27 2026 (accessed 2026-07-13)
  • cnbc.com/2026/04/27/meta-manus-china-blocks-acquisi… — CNBC: China's NDRC ordered Meta to unwind $2B Manus acquisition; Manus had passed $100M ARR in Dec 2025 (accessed 2026-07-13)
  • manus.im/docs/integrations/manus-api — Official API docs: RESTful API for task management, file handling, multi-turn conversations, webhooks; 'complete AI agent' not narrow text generation (accessed 2026-07-13)
  • awesomeagents.ai/reviews/review-manus — Independent 90-day review: GAIA scores 86.5%/70.1%/57.7% beating OpenAI; multi-agent architecture; highs impressive, lows alarming; Meta acquisition changes everything (accessed 2026-07-13)
  • aspirii.com/manus-ai-2026-review — Independent review: multi-agent system, multi-model routing (Claude + Qwen fine-tunes), credit system opaque, long-context coherence drops after 1 hour, Pro Extended $200/mo hard to justify (accessed 2026-07-13)
  • barie.ai/blog/manus-ai-review-2026-features-limitat… — Independent review: 2M+ waitlist, reliability degrades at scale, no SOC 2 or GDPR at review time, server instability, compliance unresolved (accessed 2026-07-13)
  • manus.im/blog/manus-my-computer-desktop — Official blog: Manus Desktop 'My Computer' feature launched Mar 2026, bridges cloud sandbox to local machine, CLI execution, local file access, available macOS/Windows (accessed 2026-07-13)