Data as of October 16, 2026 · How the score is built
Released Oct 15, 2026256K contextTencent Hy-Memory-30B-A3B model card
Hy-Memory-30B-A3B
Decision readingHy-Memory-30B-A3B is tracked, but not publicly ranked yet. The profile exposes 0 sourced benchmark rows and leaves unsupported fields blank until a published record exists.
Released Oct 15, 2026 — see all recent releases
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
30b-a3b · 30B-A3B
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- Weights model ID
- tencent/Hy-Memory-30B-A3BTencent Hy-Memory-30B-A3B model card
- Context window
- 256KTencent Hy-Memory-30B-A3B config
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- textTencent Hy-Memory-30B-A3B model card
- Output modalities
- textTencent Hy-Memory-30B-A3B model card
- Parameters
- Not sourced yet
- Availability
- Hugging Face · self-hosted inferenceTencent Hy-Memory-30B-A3B model card
- Cloud regions
- Not tracked yet
- Lifecycle
- activeTencent Hy-Memory-30B-A3B model card
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Not documented in the pricing recordHy-Memory technical report (dated October 15, 2026)
- Self-host
- Hy-Memory-30B-A3B needs ~24GB VRAM (1× NVIDIA RTX 4090 (24GB)).
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
We track Hy-Memory-30B-A3B, but no weighted text-model benchmark result is published on the site yet. This page shows the metadata and separate protocol evidence we can verify now; a BenchLM score will appear only if compatible public evaluations land.
Hy-Memory-30B-A3B is a open weight model with a 256K context window. No explicit reasoning mode is documented in this profile.
Open Apache 2.0 weights on Hugging Face for self-hosted inference, with the framework code and evaluation utilities on GitHub; Tencent publishes no first-party hosted endpoint or token price for this checkpoint.
Tencent Hunyuan released Hy-Memory-30B-A3B on October 15, 2026 under Apache 2.0 with a technical report: a 30B-total, about 3B-active mixture-of-experts model on the Hy3 architecture (48 layers, 128 experts with 8 active, 262,144-position context) trained as a long-term-memory writer. It replaces the general-purpose model on the write path of the Mem0, MemOS, Hy-Memory and held-out EverMemOS frameworks, deciding what to retain and how to update stored state, and is served without a chain-of-thought mode: the model card runs it with reasoning_effort set to no_think. Tencent reports a 55.49 macro average across the four harnesses on fixed 20 percent benchmark subsamples of LoCoMo, LongMemEval, PersonaMem-v2, Travel SPS, Shopping PS and Code Memory Bench, ahead of the general-purpose writers it compares against. Those harness-conditioned, subsampled memory evaluations have no lane in the catalog, so the row stays unscored and unranked.
The profile has no source-displayable benchmark row yet.
Last updated October 16, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Questions
How does Hy-Memory-30B-A3B perform overall in AI benchmarks?
Hy-Memory-30B-A3B does not have any source-displayable benchmark rows yet, so this profile does not assign a public score or rank. Documented specifications remain visible, while score-led charts and claims stay unavailable until a published evaluation can be attached to the exact model.
Is Hy-Memory-30B-A3B open source?
Hy-Memory-30B-A3B is an open-weight model from Tencent. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Does Hy-Memory-30B-A3B have full benchmark coverage on BenchLM?
No. Hy-Memory-30B-A3B currently has 0 source-displayable rows across 681 tracked benchmark slots. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of Hy-Memory-30B-A3B?
Hy-Memory-30B-A3B has a documented context window of 256K. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.