Reflection AI
Founded 2024 by Misha Laskin and Ioannis Antonoglou (ex-DeepMind). Pivoted from autonomous coding agents (Asimov) to open-weight frontier models: its first open-weight model Beam is a 501B-parameter MoE built for coding, reasoning and agentic workloads, offered with open-source harnesses/containers, a Platform API, and self-deployment (private cloud, air-gapped, edge).
Active · Verified 2026-10-29 · fresh as of 2026-10-29 · 25 cited facts, 0 with a third-party source · 2 recorded changes · Suggest a correction
At a glance
- Tier
- T2
- Created
- 2024 (founded) vendor fix
- Latest release
- not applicable — orchestration platform, not memory product fix
- License
- not applicable — not OSS fix
- GitHub
- not applicable — no GitHub repo fix
- Pricing
- Beam open weights free on Hugging Face; hosted Beam via Reflection's self-serve API Platform (API keys); no public per-token prices on the homepage vendor fix
- Freshness
- 2026-10-29 fix
- Layer
- agent-infra repo fix
- Licensing
- commercial vendor fix
- Protocols
- openai-compatible vendor fix
When to use
Optimised for: autonomous coding research agents; long-horizon software-engineering tasks
Anti-fit: not applicable — orchestration platform, not memory product
Pros & cons
Pros
Strongest ex-DeepMind RL pedigree in coding-agent space; $130M at $555M val is largest among 2024-vintage coding-agent labs; differentiated by autonomous research focus (Asimov).
Cons
Pre-product launch (private beta 2025); Devin / Cursor / Cognition / Anthropic Claude Code already established; small team vs commercial peers.
Claims & capabilities
$130M Series A Mar-2025 at $555M valuation (Lightspeed + CRV); founders ex-DeepMind AlphaGo + Gemini RLHF team; Asimov code-research agent product
Technical surface
- API surface
- not applicable — orchestration platform, not memory product
- Backend storage
- not applicable — orchestration platform, not memory product
- Deployment
- Both (self-hosted + managed cloud) — Beam open weights deployable in own environment, or via Reflection Platform API
- Embedding model
- not applicable — not a memory product
- Multi-tenancy
- not applicable — orchestration platform, not memory product
- MCP
- not applicable — orchestration platform, not memory product
- A2A
- not applicable — orchestration platform, not memory product
- OpenTelemetry
- not applicable — orchestration platform, not memory product
Alternatives to Reflection AI → live options ranked by shared design, licence, hosting and MCP support
Similar systems
Other multi-agent orchestration platforms in the catalog, ranked by inbound references.
- MultiOn (now AGI, Inc.) T2
Web-browsing agent API platform (MultiOn Agent API, docs still labelled Agent V1 Beta). Founded 2023 by Div Garg. As of 2026-10 www.multion.ai redirects to AGI, Inc. (theagi.company), which now markets AGI-0, an on-device action model and agent that operates Android apps, plus a developer Platform; the multion Python SDK has had no release since 2024-10.
- Steamship T3
YC W21 startup providing managed hosting + deployment for LLM agents — 'one-click deploy' for LangChain / agent code. Founded by Ted Benson (ex-Google Brain). Pre-seed + seed funded YC + Twin Ventures. Positioned as Heroku for AI agents — but the agent-platform space crowded post-2023, and Steamship's commercial momentum has slowed vs Vercel AI / Modal / Bedrock. As of 2026-10-29 www.steamship.com returns 404 and docs.steamship.com serves an expired TLS certificate; the last Python SDK release was v2.17.34 (2024-01-24); the python-client repo is not archived.
- Adept ACT T3
Founded 2022 by David Luan and fellow ex-OpenAI / Google researchers. Built ACT-1 / ACT-2 multimodal action transformers for computer-use. In June 2024 Amazon hired Luan and co-founders (Odena, Nye, Elsen, Szot) into Amazon AGI and licensed Adept's technology (acqui-hire, not a full acquisition); Adept continues with former head of engineering Zach Brock as CEO, refocused on agentic AI solutions. Important historical entry — ACT models inspired Anthropic Computer Use + OpenAI Operator.
- CrewAI Enterprise (now CrewAI AMP) T2
Commercial Enterprise tier of the CrewAI OSS multi-agent framework. Adds managed cloud runtime, RBAC, audit logging, SLA support, on-prem option. OSS CrewAI is in the Framework-embedded memory section already; this row covers the commercial cloud platform separately. $18M Series A Oct-2024 (Insight Partners). 30k+ companies have deployed CrewAI agents in prod (vendor-stated).
- Imbue (formerly Generally Intelligent) T3
Founded 2017 as Generally Intelligent; rebranded Imbue 2023. Founders Kanjun Qiu + Josh Albrecht. $200M Series B 2023 ($1B valuation). Trained a 70B reasoning model, then shifted to user-owned AI tools: imbue.com now lists Studio (personal AI operating system, waitlist), Bouncer, Cloud in a Bottle, Sculptor, Vet (AI code review) and mngr (CLI for managing coding agents), with open-source repos under github.com/imbue-ai.
- InstructLab (Red Hat / IBM) T2
Red Hat / IBM-stewarded open-source framework for community-driven LLM alignment + fine-tuning (LAB method, IBM Research 2024; Granite models). Apache 2.0. Main CLI repo archived 2026-04-23 after a 2025 refactor into separate libraries — synthetic data generation in SDG Hub and training in Training Hub (Red-Hat-AI-Innovation-Team). Not strictly a multi-agent framework.
Sources (24 cited facts, 0 third-party)
Each fact links to the page it came from; the badge says who published it (how to read sources).
- What it is Autonomous coding-agent foundation lab (Misha Laskin / Ioannis Antonoglou) vendor
- What it does & what's distinct Founded 2024 by Misha Laskin and Ioannis Antonoglou (ex-DeepMind). Pivoted from autonomo… vendor
- Claims / benefits $130M Series A Mar-2025 at $555M valuation (Lightspeed + CRV); founders ex-DeepMind Alph… vendor
- Created 2024 (founded) vendor
- Funding $130M Series A Mar-2025 ($555M val; Lightspeed + CRV; Sequoia, Nvidia investors) vendor
- Pricing Beam open weights free on Hugging Face; hosted Beam via Reflection's self-serve API Plat… vendor
- Deployment Both (self-hosted + managed cloud) — Beam open weights deployable in own environment, or… vendor
- HQ San Francisco, US vendor
- Founders / pedigree Misha Laskin (CEO; ex-DeepMind, AlphaGo + Gemini RLHF); Ioannis Antonoglou (CTO; ex-Deep… vendor
- Agent abstraction autonomous coding agent (long-horizon code research) vendor
- LLM lock multi-vendor (uses Claude + GPT internally per reports) vendor
- Optimised for autonomous coding research agents; long-horizon software-engineering tasks vendor
- Adjacent infrastructure DeepMind alumni network; RL research community; competes with Cognition Devin, Cursor, C… vendor
- Pros Strongest ex-DeepMind RL pedigree in coding-agent space; $130M at $555M val is largest a… vendor
- Cons Pre-product launch (private beta 2025); Devin / Cursor / Cognition / Anthropic Claude Co… vendor
- Project & sources reflection.ai/ vendor
- cost-pricing-model subscription vendor
- cost-last-verified 2026-06-29 vendor
- architecture-deployment-locus hybrid vendor
- licensing-bucket commercial vendor
- positioning-platform-vs-product platform vendor
- taxonomy-three-layer-rollup agent-infra repo
- sdk-languages rest vendor
- protocol-support openai-compatible vendor
What changed
Changes to this record's lifecycle, licence, release, pricing and hosting facts, from the catalog's git history. Subscribe (Atom)
- 2026-10-29
- Pricing “Private beta; enterprise sales early” → “Beam open weights free on Hugging Face; hosted Beam via Reflection's self-serve API Platform (API keys); no public per-token prices on the homepage” source
- 2026-10-27
- Deployment “Managed cloud (private beta)” → “Both (self-hosted + managed cloud) — Beam open weights deployable in own environment, or via Reflection Platform API” source