Open-Weight Frontier Releases as Research Substrate

Using released weights as the base for downstream research.

Tier Grades

Tier A: Frontier AI labs

feasible

high confidence

OpenAI, NVIDIA, Mistral and Alibaba are observed releasing open weights.

Tier B: Indonesian AI labs

feasible

high confidence

Sahabat-AI and SEA-LION were built on open bases.

Tier C: Institution units such as ITB

feasible

high confidence

gpt-oss-120b fits one 80GB GPU, and the compute divide paper notes academia shifting toward industry open models.

Tier D: Organisations such as SEMAI (PoT, PoC, PoV)

feasible

high confidence

Permissive licences (Apache 2.0) and single-GPU footprints make open weights the practical base for proofs of concept.

Where this comes from

Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.

Anchors

Computegpt-oss-120b runs on one 80GB GPU under Apache 2.0; Magistral Small 24B Apache 2.0.
DataNemotron publishes data, techniques and weights under the NVIDIA Open Model License; Qwen3 0.6B to 235B.
Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): not sourced in the notes.
Time horizonContinuous.
Shelf-life signalsOpen models lag closed by about 14 months weighted
Pipeline roleThe hinge of the loop: Tier A releases open weights, Tier B and Tier C adapt and study them, Tier D deploys them, with a 14-month lag behind closed models.

Where this comes from

Team size is a proxy (author counts of exemplar papers), not a headcount.

Who and What

Exemplars

gpt-oss

Nemotron

Magistral Small Alibaba (Qwen)

Qwen3

Observed actors

OpenAI

NVIDIA

Mistral AI Alibaba (Qwen)

Alibaba (Qwen)

Linked models

+255

Apertus v1.5 8Bmultimodal LLMused by

Ling 3.0 Flash (Ling-3.0-flash)LLMused by

Solar Open 2LLMused by

Kimi K3multimodal LLMused by

EXAONE 4.5-33BLLMused by

Hy3LLMused by

LongCat-2.0LLMused by

Qwen-AgentWorld-35B-A3BLLMused by

GLM-5.2LLMused by

DiffusionGemma 26B A4BLLMused by

Nemotron 3 Ultra 550B A55BLLMused by

Step-3.7-Flashmultimodal LLMused by

MiniMax-M3multimodal LLMused by

Gemma 4 12Bmultimodal LLMused by

Olmo-Hybrid-7BLLMused by

MiMo-V2.5-ProLLMused by

DeepSeek-V4-FlashLLMused by

DeepSeek-V4-ProLLMused by

DeepSeek V4LLMused by

DeepSeek-V4-Flash BaseLLMused by

Qwen3.6LLMused by

GLM-5.1LLMused by

Qwen3.6-35B-A3BLLMused by

Gemma 4LLMused by

Gemma 4 26B-A4Bmultimodal LLMused by

Gemma 4 31Bmultimodal LLMused by

Qwen3.6-27Bmultimodal LLMused by

Kimi K2.6LLMused by

Qwen3.5LLMused by

Qwen3.5-35B-A3BLLMused by

Mistral Small 4multimodal LLMused by

Nemotron 3 Nano 4BLLMused by

Sarvam 105BLLMused by

Sarvam 30BLLMused by

Step-3.5-FlashLLMused by

MiniMax-M2.5LLMused by

Qwen3.5-122B-A10Bmultimodal LLMused by

Qwen3.5-397B-A17Bmultimodal LLMused by

GLM 4.7 FlashLLMused by

Falcon-H1-ArabicLLMused by

Kimi K2.5multimodal LLMused by

GPT-OSS-SwallowLLMused by

Nemotron 3.5 Lightning 30B-A3BLLMused by

Nemotron 3 SuperLLMused by

MiMo-V2-FlashLLMused by

Olmo 3.1 32B InstructLLMused by

Mistral Large 3 (2512)multimodal LLMused by

Nemotron 3 Nano 30B-A3BLLMused by

Olmo 3LLMused by

Ling-1TLLMused by

DeepSeek-V3.2-ExpLLMused by

DeepSeek-V3.1-TerminusLLMused by

Ling-flash-2.0LLMused by

Qwen3-Next-80B-A3B-InstructLLMused by

VaultGemma 1BLLMused by

Apertus 70B Instruct (Apertus-70B-Instruct-2509)LLMused by

Apertus 8B Instruct (Apertus-8B-Instruct-2509)LLMused by

LongCat-Flash (LongCat-Flash-Chat)LLMused by

DeepSeek-V3.1LLMused by

Gemma-SEA-LION-v4-27BLLMused by

Gemma-SEA-LION-v4-27B-ITLLMused by

GLM-4.5LLMused by

gpt-oss-120bLLMused by

gpt-oss-20bLLMused by

Kimi K2LLMused by

T5Gemma 2B-2B UL2LLMused by

T5Gemma 9B-9B UL2LLMused by

MedGemma 27B textLLMused by

Qwen3-0.6B-BaseLLMused by

Qwen3-4B-BaseLLMused by

Qwen3-0.6BLLMused by

Qwen3-1.7BLLMused by

Qwen3-4BLLMused by

Qwen3-8BLLMused by

Qwen3-235B-A22BLLMused by

Qwen3-30B-A3BLLMused by

Qwen3-32BLLMused by

T5GemmaLLMused by

Nemotron-H 56B baseLLMused by

Llama 4LLMused by

Llama 4 Maverickmultimodal LLMused by

Llama 4 Scoutmultimodal LLMused by

Nemotron-HLLMused by

Nemotron-H 8B baseLLMused by

OLMo 2 32B (OLMo-2-0325-32B-Instruct)LLMused by

OLMo-2-0325-32BLLMused by

gemma-3-27b-itLLMused by

Gemma-3-4B-itLLMused by

Mistral Small 3 (Mistral-Small-24B-Instruct-2501)LLMused by

InternLM3 8B instructLLMused by

Olmo 3 7BLLMused by

DeepSeek-V2.5 (Dec 2024, DeepSeek-V2.5-1210)LLMused by

DeepSeek v3LLMused by

Falcon3-7B-InstructLLMused by

Llama-3.3-70B-InstructLLMused by

Llama 3.1 Tulu 3 70BLLMused by

Llama 3.1 Tulu 3 8BLLMused by

OLMo 2LLMused by

Llama 3.2LLMused by

Llama-3.1-Nemotron-70B-InstructLLMused by

Gemma 2 2B JPN ITLLMused by

Qwen2.5-32B-InstructLLMused by

Mistral Small (Sep '24)LLMused by

Qwen2.5-0.5B-InstructLLMused by

Qwen2.5-72B-InstructLLMused by

Qwen2.5-7B-InstructLLMused by

Qwen2.5-Math-1.5BLLMused by

Qwen2.5-Math-7BLLMused by

Qwen2.5 0.5BLLMused by

Qwen2.5-14BLLMused by

Qwen2.5-32BLLMused by LLama-3-8B-Tele-itLLMused by

Qwen2-VL 2Bmultimodal LLMused by

DeepSeek-V2.5LLMused by

DataGemma RIG 27B ITLLMused by

DataGemmaLLMused by

DataGemma RAG 27B ITLLMused by

Mistral-NeMo-Minitron 8B baseLLMused by

Llama-3.1-8B-InstructLLMused by

Mistral NeMo Instruct 2407LLMused by

Gemma 2 2BLLMused by

DCLM-7BLLMused by

InternLM2.5 7B chatLLMused by

Llama-3-ELYZA-JP-8BLLMused by

Gemma 2 9BLLMused by

Nemotron-4 340BLLMused by

RecurrentGemma 9BLLMused by

Qwen2-7BLLMused by

GLM-4-9B-ChatLLMused by

Yi-1.5LLMused by

Yi-1.5 34B ChatLLMused by

Yi-1.5 9B ChatLLMused by

Falcon 2 11BLLMused by

DeepSeek-V2LLMused by

Qwen1.5-110B-ChatLLMused by

Mixtral 8x22B InstructLLMused by

RecurrentGemmaLLMused by

RecurrentGemma 2BLLMused by

Command R+ (Apr '24)LLMused by

Command R (Mar '24)LLMused by

Qwen1.5LLMused by

Qwen1.5-72BLLMused by

InternLMLLMused by

InternLM2 chat 7BLLMused by

OLMo 7B (first release)LLMused by

Mixtral 8x7B Instruct v0.1LLMused by

Mixtral 8x7B v0.1LLMused by

deepseek-llm-67b-baseLLMused by

DeepSeek LLM 67B Chat (V1)LLMused by

TuluLLMused by

Tulu 2 DPO 70BLLMused by

ChatGLM3-6BLLMused by

Qwen-14B-ChatLLMused by

Falcon-180BLLMused by

ChatGLMLLMused by

ChatGLM2-6BLLMused by

Falcon-RW-1BLLMused by

Falcon-40BLLMused by

Falcon-7B-InstructLLMused by

Falcon (first generation)LLMused by

Falcon-7BLLMused by

BLOOMZLLMused by

BLOOMZ 7B1LLMused by

OPT 66BLLMused by

BLOOMLLMused by

BLOOM 176BLLMused by

BLOOM 560MLLMused by

BLOOM 7B1LLMused by

OPT 1.3BLLMused by

OPT 125MLLMused by

OPT 6.7BLLMused by

OPTLLMused by

GPT-NeoX-20BLLMused by

QwenLLMused by

Qwen2LLMused by

Olmo-3-1125-32BLLMused by

ERNIE-4.5-0.3BLLMused by

command-a-03-2025LLMused by

Command A TranslateLLMused by

GemmaLLMused by

Gemma 2LLMused by

Gemma2-2b-ITLLMused by

Gemma2-9b-ITLLMused by

gemma2-9b-cpt-sahabatai-v1-base (Sahabat-AI v1 base)LLMused by

granite-4.1-8bLLMused by

Granite-4.1-8B-BaseLLMused by

Granite 4.2LLMused by

Ling-flash-base-2.0LLMused by

K2 Horizon 375B A23BLLMused by

GPT-OSS-Swallow-20BLLMused by

Agents-A1-4BLLMused by

EXAONELLMused by

EXAONE 4.5LLMused by

K-EXAONELLMused by Lingzhi-72B-chatLLMused by

Llama 2LLMused by

Llama 2 70BLLMused by

Llama-2-7BLLMused by

Llama 3.1 405BLLMused by

Llama-3.1-405B-InstructLLMused by

Llama 3.1 70BLLMused by

Llama 3.1 70B InstructLLMused by

Llama 3.1 8BLLMused by

Llama-3.2-1B-InstructLLMused by

Llama3-70B InstructLLMused by

Llama 3 8BLLMused by

Meta-Llama-3-8B-InstructLLMused by

Muse GlimmerLLMused by

MiniMax-MLLMused by

MiniMax-M2LLMused by

MiniMax-M2.1LLMused by

MiniMax-M2.7LLMused by

Ministral-3-3B-Base-2512LLMused by

Ministral-3-8B-Instruct-2512LLMused by Mistral-7B-v0.1LLMused by

Mistral-Small-24B-Base-2501LLMused by

Mistral-Small-3.1-24B-Base-2503LLMused by

Mixtral-8x22B-v0.1LLMused by

KimiLLMused by

Kimi K2-0905LLMused by

Motif-3-BaseLLMused by

Nemotron 3 Ultra BaseLLMused by

Nemotron 3 Ultra GenRMLLMused by

Nemotron 3 Ultra NVFP4LLMused by

Nemotron-4-340B-BaseLLMused by T5 base Indonesian summarizationLLMused by

GPT-J 6BLLMused by

GPT-NeoLLMused by

GPT-Neo 1.3BLLMused by

GPT-Neo 2.7BLLMused by

GPT-2LLMused by

GPT-2 (124M)LLMused by

GPT-2 largeLLMused by

GPT-2 mediumLLMused by

GPT-2 XLLLMused by Pythia-1BLLMused by Qwen2.5-VLmultimodal LLMused by

Apertus 1.0LLMused by

Apertus-70B-2509LLMused by

Apertus-8B-2509LLMused by

Apertus v1.5LLMused by

Falcon-E 3BLLMused by

Falcon-H1-7B-BaseLLMused by

Solar-Open-100BLLMused by

MiMo-V2.6-Pro distilled Qwen-9B variantLLMused by

MiMo-V2.6-Pro-MOPDLLMused by

MiMo-V2.6-Pro-RLLLMused by

MiMo-V2-ProLLMused by

GLMLLMused by

GLM-4 (0520)LLMused by

GLM-4.6LLMused by

GLM-4.7LLMused by

GLM-5LLMused by

GLM-5.3-FlashXLLMused by

GLM-5.3-PrimeLLMused by

Papers

BLOOM: A 176B-Parameter Open-Access Multilingual Language ModelBigScience Workshop et al., arXiv preprint, 2022 founding
OLMo: Accelerating the Science of Language ModelsDirk Groeneveld et al., arXiv preprint, 2024 state of the art
Llama 2: Open Foundation and Fine-Tuned Chat ModelsHugo Touvron et al., arXiv preprint, 2023 state of the art
Pythia: A Suite for Analyzing Large Language Models Across Training and ScalingStella Biderman et al., arXiv preprint, 2023 state of the art
On the Societal Impact of Open Foundation ModelsSayash Kapoor et al., arXiv preprint, 2024 survey

Where this comes from

The papers to read first: where the direction began, its current best, and a survey when one exists.

Sources

Qwen3 Technical Report arxiv.org https://arxiv.org/abs/2505.09388
The Compute Divide in Machine Learning: A Threat to Academic Contribution and Scrutiny? arxiv.org https://arxiv.org/abs/2401.02452v2
Open vs. closed AI: How behind are open models? epoch.ai https://epoch.ai/blog/open-models-report
Magistral | Mistral AI mistral.ai https://mistral.ai/news/magistral
NVIDIA Nemotron nvidia.com https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/
openai.com/index/introducing-gpt-oss https://openai.com/index/introducing-gpt-oss/

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Open-Weight Frontier Releases as Research Substrate", https://aurion.semai.id/topics/open-weights-as-research-substrate/, records of October 2026.

Open-Weight Frontier Releases as Research Substrate

Tier Grades

Tier A: Frontier AI labs

Tier B: Indonesian AI labs

Tier C: Institution units such as ITB

Tier D: Organisations such as SEMAI (PoT, PoC, PoV)

Anchors

Who and What

Links

Models 279

Papers

Sources