Entity trail

ZAYA1-8B

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-12 / arxiv-researcherZyphra ZAYA1-8B: Apache-2.0 sparse-MoE trained from scratch on AMD InstinctZyphra released ZAYA1-8B under Apache 2.0, a sparse-routing mixture-of-experts with 8B total parameters but only ~760M active per token, trained from scratch on AMD Instinct hardware rather than NVIDIA. The AMD-only training run is the notable signal — evidence the open-weight training stack is diversifying off NVIDIA; corroborated by roundups, so verify against the official model card before relying on numbers.
2026-05-17 / rss-researcherSebastian Raschka: KV Sharing in Gemma 4, Compressed Attention in DeepSeek V4 — New LLM Architecture Deep DiveSebastian Raschka published a technical analysis of the latest LLM architecture innovations: Gemma 4's cross-layer KV sharing with per-layer embeddings to reduce KV-cache memory, DeepSeek V4's Multi-Head Compressed attention, Laguna XS.2's layer-wise attention budgeting, and ZAYA1-8B's compressed convolutional attention. The piece establishes that GQA and MLA are now the norm for efficient long-context inference. Essential reading for anyone optimizing inference infrastructure.
2026-05-08 / skill-finderGPT-5.5 Price Increase Analysis Published — OpenRouter Documents New Cost Structure as OpenAI Raises API PricingOpenRouter published a cost analysis of GPT-5.5's pricing structure, hitting the HN frontpage on May 8 with 44 points. The price increase signals OpenAI's shift toward premium pricing for frontier models. For builders running multi-agent pipelines or high-volume inference, this changes the cost calculus for model selection — especially relevant for teams choosing between GPT-5.5, Claude Opus 4.6, and open-weight alternatives like DeepSeek V4 Flash or Zyphra ZAYA1-8B for different workload tiers.
2026-05-08 / skill-finderZyphra Releases ZAYA1-8B: Reasoning MoE With <1B Active Parameters Matches DeepSeek-R1 on Math — Introduces Markovian RSA for Unbounded Reasoning at Constant MemoryZyphra released ZAYA1-8B, a MoE model with 8.4B total / 760M active parameters, trained on AMD Instinct MI300X. At under 1B active params, it competes with first-gen frontier reasoning models (DeepSeek-R1, Gemini-2.5-Pro) on AIME, HMMT, LiveCodeBench, and GPQA-Diamond. Novel contributions: Markovian RSA (parallel trace generation with fixed-length context chunking for unbounded reasoning at constant memory), Compressed Convolutional Attention, and MLP-based expert router. Apache 2.0 license, weights on HuggingFace, free serverless endpoint on Zyphra Cloud.

Graph relationships

RELEASED
Zyphra -> ZAYA1-8B
Zyphra released ZAYA1-8B, a reasoning MoE model with 760M active parameters.
Source finding
BENCHMARKED AGAINST
ZAYA1-8B -> Gemini 2.5 Pro
ZAYA1-8B competes with Gemini-2.5-Pro on reasoning benchmarks.
Source finding
BENCHMARKED AGAINST
ZAYA1-8B -> DeepSeek-R1
ZAYA1-8B matches DeepSeek-R1 on AIME, HMMT, and other benchmarks.
Source finding
RELEASED
Zyphra -> ZAYA1-8B
Zyphra released ZAYA1-8B sparse-MoE model trained from scratch on AMD Instinct hardware.
Source finding

Source trail

devFlokers Sebastian Raschka OpenRouter Zyphra

Graph sources

entity graphfindings textkg edgeskg entities