π0.5 (Physical Intelligence)
https://www.pi.website/blog/pi05
Extends π0 with MEM layer giving the policy short-term (within-task) + long-term (cross-task, >10-min horizon) memory. Co-trained on robot teleoperation + human video + text. Hierarchical inference: VLA predicts subtasks as text tokens, then executes as action chunks.
Active · Verified 2026-10-29 · fresh as of 2026-10-29 · 34 cited facts, 5 with a third-party source · 2 recorded changes · Suggest a correction
At a glance
- Tier
- T1
- Created
- 2024 (Physical Intelligence formally founded 2024; informal work began Q3 2023; co-founded by Chelsea Finn, Sergey Levine, Karol Hausman, Brian Ichter, Lachy Groom, Adnan Esmail and Quan Vuong) third-party fix
- Latest release
- not applicable — not OSS fix
- License
- Apache-2.0 (openpi repo; repo also ships a LICENSE_GEMMA.txt for Gemma-derived terms) repo fix
- GitHub
- Physical-Intelligence/openpi — ~13.3k stars, last commit not confirmed (π0.5 base, DROID and LIBERO checkpoints released) repo fix
- Pricing
- searched not found fix
- Funding
- total ~$1.07B (Series B $600M at $5.6B led by CapitalG, 2025-11); ~$1B at ~$11B reported in talks Mar 2026, not confirmed closed third-party fix
- Freshness
- 2026-10-29 fix
- Layer
- agent-infra repo fix
- Licensing
- open-source repo fix
- Developer activity
- OpenRank 2.92 (12-mo avg 11.25) · 14 participants — 2026-09 third-party fix
Memory taxonomy
- storage
- vector
- retrieval
- attention
- persistence
- long-term
- update
- append-only
- unit
- episode
- governance
- opaque
- conflict
- append
When to use
Optimised for: real-time perception + spatial reasoning + multi-modal sensor fusion
Anti-fit: not for non-embodied / non-physical use cases
Pros & cons
Pros
Generalist robot policy with strong simulation-to-real transfer; memory is implicit in the policy network rather than explicit external store.
Cons
Implicit memory is harder to inspect or edit; embodied agent only — not generalizable to LLM agents.
Claims & capabilities
Open-world generalisation to unseen household environments. Cleans an entirely new kitchen with no environment-specific training.
Technical surface
- API surface
- searched not found
- Backend storage
- searched not found
- Deployment
- Cloud inference + on-robot deployment (websocket streaming)
- Embedding model
- searched not found
- Multi-tenancy
- searched not found
- MCP
- no MCP support advertised — vertical product, no MCP server / client integration documented
- A2A
- no A2A protocol support advertised — vertical product, no A2A integration documented
- OpenTelemetry
- no OpenTelemetry integration advertised — vendor logs/observability not publicly documented
Alternatives to π0.5 (Physical Intelligence) → live options ranked by shared design, licence, hosting and MCP support
Similar systems
Other vertical / domain-specific ai memory in the catalog, ranked by inbound references.
- Inworld AI T1
Formerly an NPC character engine with named Long-Term Memory; now sells realtime voice models (TTS-2, STT-1, speech-to-speech) and OpenAI-compatible open-model inference. Character/NPC memory is no longer the headline product.
- NVIDIA ReMEmbR T3
Builds long-horizon memory by captioning video segments with VILA, storing captions with timestamps + 3D position coordinates in MilvusDB. At query time, LLM iterates retrieval across text, time, and position modalities. Deployed on Nova Carter robot (Jetson Orin).
- Abridge T1
Clinician-assist ambient documentation. Source mapping: every AI-generated summary element traced back to the source utterance. Audit-and-trust layer over episodic memory. Built on proprietary 1.5M+ medical-encounter dataset.
- ASAPP GenerativeAgent T1
Treats memory as first-class architecture. Captures the digital footprint of every interaction; retrieves preference and history at engagement time. Public example: airline knowing a frequent flyer wants aisle seats with her son — preference-aware, not just history-lookup.
- BenevolentAI T1
Taken private: merged into Osaka Holdings (renamed BenevolentAI S.A.) and delisted from Euronext Amsterdam 2025-03-13. Target identification / drug repurposing / mechanism tracing. 85+ data sources, petabyte-scale, rebuilt every few weeks. Wet-lab results re-enter the graph and shift downstream predictions — institutional experimental memory closing a feedback loop. As of Oct 2026 benevolent.com leads with Qor, powered by BenevolentAI, a life-science intelligence service for biopharma decision makers and consultancies. Qor (qor.one) is a continuously updated decision-intelligence service built on BenevolentAI's knowledge graph and reasoning methods: it tracks evidence on an asset or indication (competitive position, rival readouts, comparable deals), source-verifies each claim and flags when an earlier conclusion stops holding. It targets biotech teams and life-science consultancies.
- Causaly T1
Agentic AI platform for life sciences: orchestrated scientific agents over a curated biomedical knowledge graph (plus Bio Graph and Pipeline Graph), an evidence-retrieval system and an enterprise data fabric over literature, trials and patents.
Sources (33 cited facts, 5 third-party)
Each fact links to the page it came from; the badge says who published it (how to read sources).
- What it is Multi-Scale Embodied Memory (short + long episodic) vendor
- What it does & what's distinct Extends π0 with MEM layer giving the policy short-term (within-task) + long-term (cross-… vendor
- Claims / benefits Open-world generalisation to unseen household environments. Cleans an entirely new kitch… vendor
- Modalities vision + sensor + text vendor
- Created 2024 (Physical Intelligence formally founded 2024; informal work began Q3 2023; co-found… third-party
- License Apache-2.0 (openpi repo; repo also ships a LICENSE_GEMMA.txt for Gemma-derived terms) repo
- GitHub Physical-Intelligence/openpi — ~13.3k stars, last commit not confirmed (π0.5 base, DROID… repo
- Funding total ~$1.07B (Series B $600M at $5.6B led by CapitalG, 2025-11); ~$1B at ~$11B reported… third-party
- Customers / scale 51-200 ppl third-party
- Data handling Trained on Open X-Embodiment + proprietary in-house datasets; open-source weights availa… repo
- Deployment Cloud inference + on-robot deployment (websocket streaming) repo
- HQ US (San Francisco) vendor
- Founders / pedigree Physical Intelligence; founders Sergey Levine (UC Berkeley robotics professor), Karol Ha… third-party
- Memory volume / scale no scale claim vendor
- Performance π0.5 (Physical Intelligence): open-world generalisation results — clean unseen kitchens … vendor
- Code/weights release Code public (openpi, Apache-2.0) + weights public (π0.5 base, DROID and LIBERO checkpoin… repo
- Versioning model-version semantics (policy checkpoints) vendor
- Namespace primitives robot-per-instance (per-robot memory) vendor
- Contradiction handling always append (no resolution) vendor
- Forgetting policy episode-based memory; long-term memory module retains landmarks; episodic memory pruned … vendor
- Validated verticals robotics (humanoid, manipulation), autonomous-driving, embodied AI research vendor
- Time to running weeks-to-months (custom robot/vehicle integration) or research-only code release vendor
- Anti-fit not for non-embodied / non-physical use cases vendor
- Optimised for real-time perception + spatial reasoning + multi-modal sensor fusion vendor
- Adjacent infrastructure requires robot / vehicle hardware + sensor stack vendor
- Pros Generalist robot policy with strong simulation-to-real transfer; memory is implicit in t… vendor
- Cons Implicit memory is harder to inspect or edit; embodied agent only — not generalizable to… vendor
- Project & sources π0.5 blog paper vendor
- architecture-deployment-locus local repo
- licensing-bucket open-source repo
- positioning-platform-vs-product platform repo
- taxonomy-three-layer-rollup agent-infra repo
- developer-activity-opendigger {"activity":28.94,"as_of":"2026-09","openrank_12mo":11.25,"openrank_current":2.92,"parti… third-party
What changed
Changes to this record's lifecycle, licence, release, pricing and hosting facts, from the catalog's git history. Subscribe (Atom)
- 2026-10-23
- License “Apache-2.0 (openpi code; model weights also under Gemma terms, LICENSE_GEMMA.txt)” → “Apache-2.0 (openpi repo; repo also ships a LICENSE_GEMMA.txt for Gemma-derived terms)” source
- 2026-10-22
- License “not applicable — not OSS” → “Apache-2.0 (openpi code; model weights also under Gemma terms, LICENSE_GEMMA.txt)” source
- Funding / ownership “$400M total $2.0B val Series A · 2024-11” → “total ~$1.07B (Series B $600M at $5.6B led by CapitalG, 2025-11); ~$1B at ~$11B reported in talks Mar 2026, not confirmed closed” source