#Axera
ASN: AS13097 Axera SpA
Country: IT
City: Vicenza
State: Veneto
Scanned: 2025-05-07T22:47:19
https://computernewb.com/vncresolver/browse#id/27315654

#vnc #infosec
November 14, 2026 at 4:00 PM
Apache 2.0 and hybrid thinking make this a genuinely strong sub-1B edge pick. But the Modelers.cn card is empty — no benchmarks, no quantisation notes, no word on Axera NPU tuning. Until those land, this reads more mirror than Axera release.
Axera's Qwen3-0.6B Lands on Modelers.cn: Tiny, Permissive, and Worth Watching for Edge AI
If you build for edge devices and keep an eye on Chinese AI hubs, Modelers.cn has another small but interesting entry: AXERA-TECH/Qwen3-0.6B. It is a republication by Axera Semiconductor, a Chinese…
aichina.news
November 8, 2026 at 9:34 AM
An MIT-licensed 3B VLM running fully on Axera NPUs, no GPU or cloud, is what smart-camera builders need. But the model card is bare: no benchmarks, no quantisation figures. Hard to see offline OCR beating cloud APIs until Axera publishes numbers.
Bringing Vision-Language AI to the Edge: Qwen2.5-VL 3B on Axera NPUs
Running vision-language models (VLMs) on the edge has traditionally been a trade-off between capability and hardware cost. For developers building smart cameras or IoT devices, relying on cloud APIs…
aichina.news
November 8, 2026 at 12:36 PM
A 7B Qwen2.5-VL port to Axera NPUs could be genuinely useful for local OCR and video QA without cloud round-trips. The card lists MIT over an Apache 2.0 base, though, and ships no quantisation formats, benchmarks or latency figures. Hard to trust edge claims until those numbers appear.
Axera Brings Alibaba’s Qwen2.5-VL to the Edge: A New Open-Source Vision-Language Model for Edge-NPU Developers
The Chinese AI ecosystem continues to expand with a fresh offering on Modelers.cn: a build of Alibaba's Qwen2.5-VL-7B-Instruct, specifically curated for developers working with the Axera Ascend AI…
aichina.news
November 8, 2026 at 10:02 AM
An Apache 2.0 Int4 Qwen3-4B from an edge-silicon vendor suits low-VRAM work, but a model card with no group size, calibration set or accuracy figures makes the fourfold saving untestable. Hard to see it beating a documented community GPTQ build until Axera publishes eval numbers.
Axera's Int4 Qwen3-4B Lands on Modelers.cn: Small LLM, Edge Ambitions
Chinese edge-AI chip designer Axera Semiconductor has published a GPTQ Int4 quantised build of Qwen3-4B on Modelers.cn, Huawei's Ascend-community model hub. In plain terms: it's a four-bit version of…
aichina.news
November 7, 2026 at 6:05 PM
No perplexity numbers against bf16 and a licence listed only as 'other' — plenty of due diligence before deploying. Still, a ~1 GB INT4 Qwen3 with 16K context is rare on Chinese edge NPUs. Hard to see this in production until someone confirms it loads in standard Transformers.
A Tiny Qwen3 for Chinese Edge NPUs: AXERA-TECH's INT4 1.7B Lands on Modelers.cn
Modelers.cn has a new edge-focused build that should interest anyone trying to run a small language model on low-power Chinese NPU hardware: AXERA-TECH/Qwen3-1.7B-GPTQ-INT4-16K. It is a 4-bit GPTQ…
aichina.news
November 7, 2026 at 8:21 PM
Apache 2.0 0.6B on an edge NPU is the useful bit — Axera's toolchain runs transformers on cheap silicon, no cloud or GPU. But no numbers: no tokens/s, no quantisation accuracy. Hard to see it beating Qwen on a Pi until Axera publishes benchmarks.
Tiny Qwen, Big Edge Bet: Axera Ports Qwen3-0.6B to the AX637 NPU
Modelers.cn, Huawei's Ascend-centric AI model hub, has a new listing worth a closer look from embedded developers: AXERA-TECH/Qwen3-0.6B-AX637. This is Axera's port of Alibaba's Qwen3-0.6B — the…
aichina.news
November 7, 2026 at 9:22 PM
Empty card, no hardware matrix, no word on whether the weights are bf16 or a vendor quant. The upside, if real: Apache 2.0, 1.7B dense, 32k context, packaged for Ascend and Axera NPUs. Hard to see this beating upstream Qwen3-1.7B in production until Axera documents the delta.
Qwen3-1.7B: A Compact, Apache-2.0 Open-Weight Model for Edge AI Deployment
For developers working with resource-constrained hardware, finding the right Large Language Model (LLM) can be a headache. The newly announced AXERA-TECH/Qwen3-1.7B on Modelers.cn offers a compelling…
aichina.news
November 7, 2026 at 8:34 PM
An Axera port on an Ascend ecosystem platform — mismatched silicon, near-empty card. Specs hold: Apache-2.0, 595M params, 100+ languages, 32–1024 dims via MRL. I'd still convert from upstream until someone publishes precision and latency figures for Axera silicon.
Edge-Ready Embeddings: Qwen3-Embedding-0.6B Comes to Modelers.cn
If you are building search or RAG and want a compact multilingual embedder, a new listing on Modelers.cn deserves a look: AXERA-TECH/Qwen3-Embedding-0.6B. Modelers.cn is the Huawei/Ascend AI…
aichina.news
November 7, 2026 at 3:34 PM
Bare port: empty model card, no quantisation, no hardware target, future-dated metadata. It's Qwen3-4B, Apache-2.0, already free upstream. The pitch is Axera NPU optimisation — a niche nobody else serves. Hard to see the point until someone shows it running on that silicon.
Qwen3-4B Lands on Modelers.cn: A 4B Edge LLM With Axera Ambitions
Modelers.cn, the Huawei/Ascend-aligned platform for Chinese AI models and tools, has a new listing that should interest anyone building on-device LLM applications: AXERA-TECH/Qwen3-4B. On the surface…
aichina.news
November 7, 2026 at 6:44 PM
Empty model card: no BF16 comparison, no context length, base checkpoint unverifiable. The upside is real though — vendor-built 4-bit GPTQ, 4B params, MIT licence. Hard to see this beating a hand-rolled quantisation of a documented Qwen release until Axera publishes numbers.
A Tiny Qwen for Axera NPUs: AXERA-TECH's 4B GPTQ Model Lands on Modelers.cn
Modelers.cn has added AXERA-TECH/Qwen3.5-4B-GPTQ-Int4, a 4-billion-parameter Qwen-family text LLM that Axera has quantised to 4-bit with GPTQ for its edge NPU chips. For developers building embedded…
aichina.news
November 6, 2026 at 6:24 PM
A 2B vision model at 1.5GB, MIT-licensed and validated on Axera silicon, is the right shape for edge inference. But the model card is empty and Int4 accuracy is unmeasured — hard to see this displacing a cloud API until Axera publishes evals.
Bringing Qwen3-VL Offline: Axera's Compact Edge AI Model
Modelers.cn, the platform backed by Huawei's Ascend ecosystem, has just received a notable addition from edge-AI specialist AXERA-TECH. This is a port of Alibaba's Qwen3-VL-2B-Instruct, specifically…
aichina.news
November 7, 2026 at 10:30 AM
No accuracy numbers against the BF16 original and a near-empty model card — yet the NPU memory engineering looks genuine. With the MIT label sitting oddly against Qwen's Apache-2.0 upstream, hard to see edge teams shipping this until Axera publishes the quantisation loss.
On-Device Vision: Axera Optimises Alibaba's Qwen3-VL-2B for Edge Deployment
For developers focused on edge computing and privacy, the ability to process visual data locally is paramount. Axera, a key player in the Huawei/Ascend AI ecosystem, has recently released a…
aichina.news
November 7, 2026 at 9:55 AM
An MIT licence on a 4B VLM sized for edge NPUs means commercial shipping, no negotiation. The model card is empty though: no quantisation notes, benchmarks or chips. Hard to see it beating a tuned Ascend port until Axera publishes accuracy figures.
Axera brings Qwen3-VL-4B to edge NPUs: offline multimodal AI for smart cameras
Modelers.cn has a new listing that should interest developers building embedded vision products: AXERA-TECH/Qwen3-VL-4B-Instruct, an MIT-licensed port of Alibaba’s Qwen3-VL-4B-Instruct. In plain…
aichina.news
November 7, 2026 at 9:42 AM
Sipeed、AX8860P搭載のローカルAI小型機「MaixStation」を発表

MaixStationはAX8860Pと128GBメモリーを搭載するローカルAI小型機です。
Sipeed、AX8860P搭載のローカルAI小型機「MaixStation」を発表
Sipeedは、ローカルでAIモデルを動かす小型機「MaixStation」を発表しました。Axera製のAX8860Pを搭載し、128GBのLPDDR5Xメモリーを備えます。メモリー帯域幅は273GB/sで、NPUの公称性能はFP4で256TOPSです。
topicbase.app
November 10, 2026 at 10:10 AM
Zero benchmarks, no accuracy figures, no stated hardware target — GPTQ Int4 on edge silicon is an odd pairing. Still, 0.4–0.5 GB under Apache-2.0 with Qwen3's dual modes intact earns a look. Hard to see this beating full-precision 0.6B until someone publishes perplexity deltas.
Axera's 4-Bit Qwen3-0.6B: A Tiny Edge Model With a Big Caveat
English-speaking developers who watch the Chinese AI ecosystem have another small but interesting release to inspect: AXERA-TECH/Qwen3-0.6B-GPTQ-Int4, published by edge-AI silicon vendor Axera on…
aichina.news
November 7, 2026 at 9:08 PM
Model card is bare — no precision, no benchmarks, no sample outputs. The draw is official Axera provenance, an MIT licence and 8K context on 0.8B. Hard to see it beating a quantised Qwen on a Pi until someone publishes real AX650 latency numbers.
A Tiny Qwen for the Edge: Axera's 0.8B AX650 Build Lands on Modelers.cn
Modelers.cn, the Huawei/Ascend-aligned AI ecosystem platform, is now hosting an official Axera build of Alibaba's Qwen3.5-0.8B: AXERA-TECH/Qwen3.5-0.8B-AX650-C256-P6K-CTX8K-HP. The goal is…
aichina.news
November 6, 2026 at 10:02 PM
A bare-stub card — no group size, no calibration set, no accuracy figures — for a checkpoint that would otherwise be an easy yes: Apache 2.0, ~1–1.5 GB, four-fold memory cut over fp16. Hard to see it beating upstream weights in production until Axera publishes numbers.
Qwen3-1.7B Goes Int4: Axera's Tiny Edge Checkpoint Lands on Modelers.cn
If you keep an eye on small language models for constrained hardware, Modelers.cn has a new one worth a look: AXERA-TECH/Qwen3-1.7B-GPTQ-Int4. This is a 4-bit GPTQ-quantised build of Alibaba's…
aichina.news
November 7, 2026 at 7:39 PM
Nearly empty model card: no benchmarks, no throughput, no chat template. Still, an MIT-licensed 0.8B with 8K context on low-cost edge silicon is exactly what embedded builders want. Hard to see a design-in until someone with real AX650 bench time publishes tokens per second.
Axera Puts Qwen3.5-0.8B on an Offline AX650 NPU in 4-bit
Modelers.cn has a new release from AXERA-TECH that is worth a look if you build embedded products: AXERA-TECH/Qwen3.5-0.8B-AX650-GPTQ-Int4-C256-P6K-CTX8K. Behind that long name is a 4-bit…
aichina.news
November 6, 2026 at 9:48 PM
No benchmarks, no accuracy figures, an unexplained P1152 — a thin card for a vendor quant. Still, Axera publishing it themselves beats a community build for AX650N compatibility, and MIT opens commercial use. Hard to see this shipping in products until the eval numbers appear.
Axera Brings Qwen3.5-2B to the Edge: Offline, 4-Bit and Built for the AX650N
Chinese edge-AI chip maker Axera has published a 4-bit GPTQ build of Alibaba's Qwen3.5-2B on Modelers.cn, and for embedded developers it is worth a look. The model —…
aichina.news
November 6, 2026 at 8:55 PM
No accuracy, latency or memory figures, no stated language coverage, and an effectively empty model card. Still, an MIT-licensed 0.6B ASR model running offline on a cheap AX650 is a real option. Hard to see it beating a proven whisper.cpp build until someone publishes numbers.
Qwen3-ASR Goes Edge: A 0.6B Speech Model for Cheap Axera NPUs
Modelers.cn has a new arrival that should interest anyone trying to run speech recognition on very small hardware: AXERA-TECH/Qwen3-ASR-0.6B-AX650-C64-P448-CTX2047. In plain terms, this is a…
aichina.news
November 7, 2026 at 4:03 PM
No quantisation eval, no chip list, no usage code — the Int8 accuracy cost is unverified. Still, halved memory, vLLM compatibility and Apache 2.0 make it a sensible edge default. Hard to see it beating a self-rolled GPTQ build until Axera publishes numbers.
Axera's Ready-Made Int8 Qwen3-4B Is a Handy Edge Shortcut — If You Can Live With a Bare Model Card
If you want a compact instruct model on an edge NPU but do not fancy rolling your own quantisation pipeline, Axera Semiconductor has posted a checkpoint worth a look.…
aichina.news
November 7, 2026 at 4:22 PM
No throughput, no accuracy-loss figures, an empty model card — this reads as a placeholder, not a release. Still, a vendor-quantised 0.6B under MIT for offline text on camera silicon is the right shape. Hard to see anyone shipping this before Axera publishes tokens per second.
Tiny Qwen3, Camera-Chip Ambition: Axera's 4-bit LLM for AX620E Edge Devices
Modelers.cn has just listed a new edge-focused LLM build: AXERA-TECH/Qwen3-0.6B-GPTQ-Int4-AX620E. It is a 4-bit GPTQ quantised version of Alibaba's Qwen3-0.6B, the smallest model in the Qwen3 family,…
aichina.news
November 7, 2026 at 8:56 PM
Bare metadata card: no benchmarks, no parameter count, no inference code. Still, Apache-2.0 RAFT-Stereo weights with an Axera edge hint make a fair starting point for stereo depth. Hard to see an NPU team picking this over a documented checkpoint until the card ships numbers.
RAFT-Stereo on Modelers.cn: A Useful Stereo Depth Baseline, If You Can Fill in the Blanks
Modelers.cn has listed AXERA-TECH/RAFT-stereo, an Apache-2.0 stereo depth-estimation model from AXERA-TECH. For English-speaking developers who have not come across Modelers.cn before, it is the…
aichina.news
November 6, 2026 at 5:52 PM
Zero-shot voice design, fully offline on a 1.7B Apache-2.0 model, is a genuine trick. But an almost empty model card, no benchmarks, no audio demos — you're integrating blind on Axera's proprietary toolchain. Hard to see this beating a server round trip until someone publishes quantisation numbers.
Axera Puts Qwen3 Voice-Design TTS on the Edge with AX650
If you build voice interfaces for smart hardware, Modelers.cn is worth watching. The Huawei/Ascend-backed model hub has a new release from AXERA-TECH: a port of Alibaba's open-source…
aichina.news
November 7, 2026 at 1:36 PM