Open-Weight Lineage
109 records · 1 notes · records tagged with this as a related axis are included.
Another family tree, built by models whose weights were released
Models whose weights you can download, run and modify yourself get used in a completely different way. This axis follows that lineage — which models were released under which terms, which derivative models branched off from them, and what the word "open" actually allows under each license. Being released doesn't automatically mean being free, so license terms are checked against the original text, not the announcement wording.
Company-reported, it covers 34 language and script combinations with 2.601 trillion tokens in native scripts plus 224.6 billion in Latin script, and includes 438.4 billion restored tokens kept only where the quality scor
It is the first model from a new US lab. Company-reported scores were SWEBench Verified 81.6%, GPQA Diamond 90.6%, AIME 2026 97.7% and Terminal Bench v2.1 80.1% (on BF16), and it ships in three versions, BF16, FP8 and NV
Company-reported, the 209M model ranks No. 1 in the pretrained zero-shot categories of GIFT-Eval, TIME and fev-bench, and the 28M model beat Chronos-2, which is about four times larger. On an 8-thread server CPU the 5M m
Because these are activated LoRAs, the KV cache built by the base model can be reused for the judgment that follows, and IBM said GPU memory with all three components added stays close to that of the base model alone. It
It shipped two versions, a 16-step distilled one and a 50-step base one, together with FP8 and NVFP4 quantized weights and a DreamBooth LoRA training script. Only a speed figure was given — the distilled version makes a
Meta claims it is the most capable US open-weights model to date. It adds safety post-training for standalone use on top of the same base model as Muse Spark 1.3, and ships with SGLang and vLLM optimizations and Helion k
Company-reported MTEB Code rose from 68.76 to 78.68; context is 8K (4 times its predecessor), and MRL lets the 768 dimensions shrink to 128, cutting storage by up to 6 times. Quantized, it runs on a Pixel 11 Pro in about
The point at which deep-research-style capabilities came down to an 8B open-weight model. Using citation accuracy and recall as evaluation axes matters, because in report generation the problem is not sentence quality bu
The first case of a competing lab putting numbers on what leaks out when open weights are released without safety measures. The debate in the open-weight lineage (axis ⑨) shifts from a "performance gap" to a "safeguard g
A path has opened for agent runtimes to move from vendor products to foundation-owned standard infrastructure. The key is defining the harness, model and sandbox as swappable slots: if this separation takes hold, the cos
A case of pushing the diagnosis that the bottleneck for agent workloads has shifted from compute to KV-cache storage and bandwidth through the entire model design. Using different active parameters for prefill and decode
A case of a large open-weight MoE coming down from general purpose to the vertical task of translation, squarely targeting commercial translation services.
Shows that a fully end-to-end unified architecture with neither a separate vision encoder nor a VAE has reached practical performance, suggesting that the standard design of unified multimodal models may change.
A case of releasing, down to the weights and data, a result that reached the IMO gold-medal range through pure natural-language reasoning without a formal checker — greatly widening the reproducibility of top-level mathe
A turning point at which Europe's open-weight, sovereign AI camp secured capital to rival US big tech, notable for the direct participation of semiconductor supply-chain companies such as Samsung Electronics, NVIDIA and
Overturns the conventional wisdom that "errors in recurrent state accumulate over long contexts", arguing that the recurrent half of a hybrid LLM is in fact the easier half to quantize, and offers four mechanistic explan
A case of the diffusion language model family crossing over into image generation, adding to the accumulated evidence for the observation topic "diffusion language models". In the authors' own reporting, it claims 53.53
MSL keeps to its closed-API route while putting agent efficiency (fewer tool calls and tokens) forward as its headline metric
A case of a frontier open-weight model disclosing the rationale for its architecture choices ablation by ablation. In the authors' own reporting, loss and downstream accuracy do not always move together, and combining th
The gap between open and closed models narrows to 3 points
Open-weight models came out in a cluster within a single week, along with the claim that a 27B-class model runs in the 200-tokens-per-second range on a single consumer GPU while coming close to top commercial models. In
The first open weights since Llama 4, released under Apache 2.0; an on-device agent at under 20GB in 4-bit
A practice in which the promise to open up a top model and its actual release come apart. It was pointed out at the same time that Kimi K3 had done the same — grounds for recording the announcement date and the weight re
While the top model moves to its own license, the small model stays on Apache 2.0 — a two-track licensing split by scale
The top open-weight model reverts to a custom license rather than Apache. Qwen3.8-Max is a closed commercial version built on it
More than 3 times the 236B of phase 1. The switch to Apache 2.0 removes commercial restrictions
An open-weight model in the 3-trillion-parameter class. A prime example of the pattern in which scale grows while licenses narrow
A phase-2 result of the independent foundation model program
A release of the harness and agent scaffolding, not of weights. The license is not specified in the announcement
An open-weight coding model that puts token efficiency up front
A small open-weight coding agent model
Released a 200B-class MoE under Apache 2.0 while lowering the bar for on-premises deployment to 1–2 GPUs
Re-entered the frontier while keeping MIT
A dense 27B beat its 397B MoE predecessor, and the center of gravity of truly open models moved down to the 30B class
An open-weight model entered the 80% range on SWE-Bench Verified. No independent party checking the result is named
As LG's first open-weight vision-language model, it marks the point where Korea's open-weight route expanded beyond text
The replacement of the Llama brand and the shift to closed
The point at which Meta's two-track strategy was settled: the frontier model closed, with only separate small models open
On-device AI on Android evolved into a system in which a system service (AICore) distributes open models directly
The opening signal of the US camp's return to truly open releases
A lab centered on text LLMs released speech recognition under a fully open license
A move toward mechanically checking AI-generated code against formal specifications. Shifted the evaluation axis from competition math to contributions to real formalization projects
Applies early-fusion multimodal training to a large Apache 2.0 MoE. Plus remains closed, so open and closed run in parallel
Open-weight distribution is endorsed as a business model fit for a public listing
Optimized for agentic AI. The successor in the Kanana-1.5 lineage
Released a 675B-class frontier model under Apache 2.0. The largest fully open release in the West as of late 2025
Open-weight reasoning models; DeepSeek claimed V3.2-Speciale reached gold-level results at IMO, IOI and other 2025 competitions
A US community team fine-tuning on top of a Chinese base model — a symbolic scene of shifting leadership
The moment large Korean open-weight models appeared all at once
An open-weight reasoning agent trained to interleave thinking with tool calls; Moonshot claims it sustains 200–300 sequential tool calls without human intervention
An approach that benchmarks lower refusal rates as an alignment metric in their own right. The open community formalizes an alternative axis to safety alignment
A turning point where the release policy retreated from Apache 2.0 to a custom community license. Because of its non-compete clause, opinions differ on whether it meets the open-source definition
Billed by DeepSeek as its first step toward the agent era; the API's deepseek-chat and deepseek-reasoner became the non-thinking and thinking modes of one model, merging what had been separate V3 and R1 offerings
OpenAI's first open-weight language models since GPT-2; 120b (117B total, 5.1B active per token) runs on a single 80 GB GPU, and 20b (21B total, 3.6B active) within 16 GB of memory
Unlike A.X 4.0, which continued pretraining open-source models on Korean data, it was designed and trained entirely in-house; SKT's open models now numbered four, two A.X 3.1 and two A.X 4.0
Alibaba claimed state-of-the-art results among open models in agentic coding, browser use and tool use, and open-sourced the command-line coding tool Qwen Code alongside it
A hybrid that integrates reasoning and agent capabilities (MCP, function calling)
The first large open-weight release by a telecom operator
The first 1T-class open-weight model, and a new type of license with revenue and scale thresholds
A turning point for the Llama line; Behemoth's release postponed
A milestone at which small, on-device-class models began to combine multimodality with competitive benchmark scores
The first Kanana release under a license that allows commercial use (Apache 2.0)
A case of a Gulf state framing an open-weight model in its own language as the language of sovereignty
A strategy of covering every size class with an unrestricted license overtook Llama in the number of derivative models on HF
Naver's first open-weight release
A turning point for trust in the Llama brand
LG's open-weight line specialized for math, science and coding reasoning; company-reported scores for the 32B were 94.5 on the 2025 CSAT math section and 90.0 on AIME 2024
The maturing of the small, on-device route
The full release following QwQ-32B-Preview; Alibaba claims performance comparable to DeepSeek-R1 (671B total, 37B active) at 32B
The first open weights in the Kanana line, released alongside a report on training the 2.1B–32.5B Kanana lineup
Mistral recommitted to Apache 2.0 for its general-purpose models while moving away from MRL-licensed ones; company-reported over 81% on MMLU at 150 tokens/s
Reasoning models were immediately reproduced as open weights, and 2 Chinese labs declared their pursuit of the frontier on the same day
A follow-up four months after the EXAONE 3.0 7.8B release that expanded to three sizes while keeping a non-commercial license
An update to the Llama 3 line that shipped a single 70B instruction-tuned model, with no separate pretrained release
With an MIT license and extreme cost efficiency, the Chinese camp took the most open position
Established the grounds for ruling that most of Llama, Gemma and Qwen are not open source
The first confirmed case of Japanese state funding actually producing a domestic 100B-class model
The 11B and 90B were the first Llama models to support image input, released alongside 1B and 3B models aimed at on-device use
All open models except the 3B and 72B were released under Apache 2.0, and the 14B and 32B sizes missing from the previous generation returned
A preview released as weights plus free API access ahead of the full release planned for November; Upstage claimed it competes with multi-GPU models such as Llama 3.1 70B
The first case of the community actually fine-tuning the 405B model
The first open release of an in-house LLM by a major Korean company. But its research-only license also revealed the conservatism of the Korean approach to openness
Published the weights of a 123B flagship designed for single-node inference, limited to research and non-commercial use, with a commercial license required for self-deployment
Frontier-class weights were released for the first time, and the license allowed training on the model's outputs
The second Gemma generation; Google claimed the 27B competes with models more than twice its size and the 9B outperforms Llama 3 8B
An open-weight code model that DeepSeek claimed matched GPT-4 Turbo on code-specific tasks
An early case of a major Japanese company publicly releasing a pretrained model in its own language, and a sign of telecom capital entering homegrown LLMs
Publicly demonstrated that safety alignment can be undone without retraining; technical grounds for the open-weight safety debate
Only the 72B kept the original Qianwen License while the other four moved to Apache 2.0, the release where Qwen's licensing began to open up
Introduced a non-production license that publishes weights for research and testing only, with commercial use requiring a separate license
The 8B delivered the previous generation's 70B-class performance, raising the bar for practical local use
Mistral's larger MoE follow-up to Mixtral 8x7B, again under Apache 2.0
The starting point of a Japanese telecom approach that, instead of following the parameter-count race, set its course toward lightweight models specialized in the national language
The largest Apache 2.0 open-weight MoE at the time. But as a raw checkpoint without fine-tuning for dialogue, its practical usability was limited
An API flagship launched with Microsoft Azure as Mistral's first distribution partner; Mistral claimed it ranked second only to GPT-4 among models available through an API
Big Tech entered the small open-weight field, but drew criticism that the license was not Apache
A model from a Korean startup that, per Upstage, topped the Hugging Face Open LLM Leaderboard at release; the 74.2 average belonged to the instruction-tuned SOLAR-10.7B-Instruct, while the base model scored 66.04
Made Sparse MoE a standard architecture of the open ecosystem and imprinted the concept of total versus active parameters
A general-purpose DeepSeek model released down to its intermediate checkpoints; DeepSeek claimed the 67B Base outperformed Llama 2 70B Base in reasoning, coding, math and Chinese comprehension
A truly unrestricted license and a challenger from Europe; later the base for thousands of fine-tunes
A code-specialized line trained on 500B tokens of code and code-related data on top of Llama 2; a 70B version was added on January 29, 2024
The start of Korea's first lineage of large commercial LLMs
Alibaba's first open-weight language model; the original release was trained on 2.2T tokens, raised to 2.4T in an update on September 25, 2023
Popularized open weights, and the starting point of the debate over whether it is open source
Established at once the practice of training on synthetic and leaked conversations and the culture of LLM-as-judge evaluation
Demonstrated that instruction-following can be replicated for a small sum
Proved that controlled distribution is physically impossible; the real starting point of the open-weight ecosystem
The first large attempt to distribute frontier-class weights after vetting
Presented by Naver as Korea's first hyperscale AI and a means to AI sovereignty; a closed model later succeeded by HyperCLOVA X