Updated Oct 29

The Open-Weight Frontier

The gap between the best downloadable model and the best closed one more than doubled on paper this week, because the ruler changed. On 27 October Artificial Analysis rebuilt its Intelligence Index around private and harder tests. The best models with published weights, DeepSeek's V4.1 Flash 1017 and Zhipu's GLM-5.3, score 32 on it, 26 points behind the leader; a day later Google's closed Gemini 4 Carbon took the top at 59. On the old version the gap was twelve. The analyst Nathan Lambert says the old index flattered open models; Epoch AI's cyber measure, a different method, puts the best open model six to seven months behind.

Two labs published weights they had promised. MiniMax released M3.1 Flash on 23 October, four days after launch; it scores 29. Mistral released Large 4, about a trillion parameters, on 29 October, inside its end-of-month promise, under a licence that charges hosting businesses; it scores 26.

Moonshot's Kimi K3.1 Pro scored fourth of all models at its 23 October launch, and the index said its weights were expected soon after. They have not appeared. Meta's Muse Spark Open, at 25, is still the best American open model.

New Oct 29

The whole story5 chapters

The state of it — as of October 29, 2026

27 points

best downloadable model (32) behind the best closed one (59), Artificial Analysis index v5

best American open-weight model, Meta's Muse Spark Open, on the same index

6–7 months

lag of the best open model behind closed ones on Epoch's cyber index

Today

All new

Oct 29, 2026

Mistral Publishes Large 4's Weights, Inside Its October Promise

Mistral published the weights of Large 4, a model of about a trillion parameters with 49 billion active and a million-token context, on 29 October, having promised them for the end of the month when the model went on sale in early October. The licence is Mistral's own modified MIT: free to use and modify, but businesses selling it as a hosted service need a separate commercial licence. Artificial Analysis scores it 26 on its rebuilt index, seventh among open models. Mistral kept a specialised cyber variant out of the release. The full weights run on eight H200 GPUs.

Read

Mistral AIHugging FaceThe Next WebArtificial AnalysisLindyHugging FaceArtificial Analysis

Oct 28, 2026

Open models' demise from this seems greatly exaggerated, but I do think the previous index was flattering the open weight models too strongly.

— Nathan Lambert

Not yet checked

A judgement about a measure, not yet checkable. On the rebuilt index the best released open model fell from 46 (twelve behind) to 32 (26 behind), per Artificial Analysis.

Read

Nathan LambertArtificial Analysis

Oct 27, 2026

On a Rebuilt Index, the Best Downloadable Model Falls 26 Points Behind

Artificial Analysis rebuilt its Intelligence Index, moving 60% of it onto private or held-out tests and adding harder coding and science tasks. Open-weight models fell further than closed ones. The best with published weights, DeepSeek V4.1 Flash 1017 and Zhipu's GLM-5.3, score 32, 26 points behind the leader at 58 and below a small closed model at 36; on the previous version the gap was twelve. Meta's Muse Spark Open, the best American open model, scores 25. Part of the widening is the change of measure itself. A day later Google's closed Gemini 4 Carbon took the top at 59, putting the gap at 27.

Read

Artificial AnalysisArtificial AnalysisArtificial Analysis

Earlier this month

NewOct 23, 2026

MiniMax Publishes M3.1 Flash's Weights, Four Days After Launch

MiniMax published the weights of M3.1 Flash on Hugging Face on 23 October, four days after the model went on sale as a service, having said only that weights would follow 'later'. It is a 428-billion-parameter model with 23 billion active per token, taking text, image and video input, with a context of up to a million tokens. On Artificial Analysis's rebuilt index it scores 29, in a group of three just behind the open tier's top three. It ships under MiniMax's own community licence, which allows commercial use with restrictions.

Read

Hugging FaceTechNodeArtificial AnalysisArtificial Analysis

NewOct 23, 2026

Kimi K3.1 Pro Scores Fourth Overall, but Its Weights Are Not Out

Moonshot launched Kimi K3.1 Pro, a 2.8-trillion-parameter model with 104 billion active, as a paid service at $2 per million input tokens and $10 output. Artificial Analysis scored it 55 at launch, fourth of all models, and said the weights were not published but expected soon after launch; once out, it said, K3.1 Pro would lead the open tier. Moonshot's own documentation covers API access only. Six days later the index listed it as proprietary, its weights not publicly available; on the rebuilt version it scores 46.

Read

Artificial AnalysisMoonshot AIArtificial AnalysisArtificial AnalysisArtificial Analysis

Oct 17, 2026

In reality, this was much more likely a safety tune of Muse Spark 1.2, which would only be a 1 point drop on Artificial Analysis to shore up safety behaviors.

— Nathan Lambert

Not yet checked

Still unverified. Meta's own model card says only that the open version "has additional safety post-training" for use without system guardrails; it names no Muse Spark version it was built from.

Read

Nathan LambertHugging Face

Oct 17, 2026

DeepSeek's October Update Ties for Second-Best Model Anyone Can Download

DeepSeek published the weights of V4.1 Flash 1017, an October update to its V4.1 Flash, on Hugging Face under the MIT licence: 552 billion parameters, 16 billion active per token. Artificial Analysis scores it 45, six points above the September version and level with Zhipu's GLM-5.3 for second among open-weight models, a point behind Xiaomi's MiMo-V2.6-Pro. The best closed model still scores 58. It stays among the cheapest to run: $0.30 per million input tokens and $1.20 output on DeepSeek's own service, and from $0.15 input across the 17 hosts the index tracks.

Read

Hugging FaceArtificial AnalysisArtificial AnalysisArtificial AnalysisArtificial Analysis

Oct 16, 2026

Reflection Ships Beam's Weights, and It Scores 30

Reflection AI released the weights of Beam, a 501-billion-parameter model with 23 billion active, on Hugging Face under the Apache 2.0 licence as it had promised. Artificial Analysis, which tested the model before release as one of Reflection's partners, scores it 30: second among American open-weight models behind Meta's Muse Spark Open at 39, and sixteen points behind the best Chinese open model. It used fewer tokens to run the index than comparable open models. Four providers serve it, at a median of $0.70 per million input tokens and $2.10 per million output, according to Artificial Analysis.

Read

Reflection AIArtificial AnalysisArtificial AnalysisThreat FrontierArtificial AnalysisNebius

NewOct 16, 2026

On Epoch's Cyber Measure, the Best Open Model Is Six to Seven Months Behind

Epoch AI's cyber-capabilities index puts the best open-weight model, Zhipu's GLM-5.3, at 160: above OpenAI's GPT-5.4, released in March, and below GPT-5.5 and Claude Mythos Preview, both from April. Epoch concludes that the open cyber frontier sits roughly where the closed frontier stood six to seven months earlier. The best closed models now score 171.

Read

Epoch AIEpoch AI

Oct 15, 2026

Meta Opens an 816-Billion-Parameter Model, on Its Own Terms

Meta published the weights of Muse Spark Open, a mixture-of-experts model with 816 billion parameters, 42 billion active per token. Artificial Analysis scores it 39, the highest of any American open-weight model, but seven points behind Xiaomi's MiMo-V2.6-Pro and nineteen behind the top closed model. The licence is Meta's own rather than an open-source one: products must display "Built with Muse Spark Open", hosting services with more than $20 million in revenue must sign a separate agreement with Meta, and military use is barred. Running it takes at least 700 gigabytes of GPU memory, by Meta's count.

Read

Hugging FaceHugging FaceArtificial AnalysisArtificial AnalysisArtificial AnalysisThe GuardianTrending TopicsHugging Face

Oct 15, 2026

is the most capable US open-weight model to date

— Alexandr Wang True

Checked: true

Holds on the independent index: Artificial Analysis scores Muse Spark Open 39, above every other American open-weight model it lists, the next being Reflection's Beam at 30. It is still seven points behind the best Chinese open model.

Read

The GuardianArtificial AnalysisArtificial AnalysisArtificial Analysis

Oct 15, 2026

StepFun Publishes Step 5 Preview, Third-Best Model Anyone Can Download

StepFun published the weights of Step 5 Preview, a 600-billion-parameter model with 27 billion active, on Hugging Face and ModelScope. Its API had been running since September. Artificial Analysis scores it 44, third among open-weight models behind Xiaomi's MiMo-V2.6-Pro at 46 and Zhipu's GLM-5.3 at 45, and fourteen points below the best closed model. The licence is StepFun's own: Apache-style, with added attribution and commercial-licence conditions. Running it takes a data-centre node of four B300 or eight H200 GPUs.

Read

Hugging FaceArtificial AnalysisArtificial AnalysisMarkTechPost

Oct 6, 2026

An American Lab Announces an Open-Weight Challenger, Weights to Follow

Reflection AI, an American lab founded in 2024 by former Google DeepMind researchers, announced Beam, its first model: 501 billion parameters, 23 billion active per token, text only. The weights were not released with the announcement; access is by waitlist while testing finishes. No independent benchmark exists, and Beam is not on the Artificial Analysis index. On Reflection's own published scores it trails the leading Chinese open models — 44.4 on DeepSWE v1.1 against 68.0 for Kimi K3, 61.0 for GLM-5.3 and 74.2 for DeepSeek V4.1 Flash.

Read

This quarter

Sep 21, 2026

Xiaomi Now Publishes the Best Model Anyone Can Download

Xiaomi released MiMo-V2.6-Pro, a one-trillion-parameter model with 42 billion parameters active per token, under an MIT licence. Artificial Analysis scores it 46 on Intelligence Index v4.3.2, the highest of the 82 open-weight models it ranks, ahead of Zhipu's GLM-5.3 at 45 and Moonshot's Kimi K3 at 44, at $0.43 and $0.87 per million tokens. The closed leader a day later, Claude Opus 5.5, scored 58: a twelve-point margin. Before Opus 5.5 the top score on the same scale was 53 and the best open model 45, a margin of eight.

Read

Artificial AnalysisArtificial AnalysisArtificial AnalysisArtificial AnalysisArtificial AnalysisArtificial AnalysisTrending Topics

Sep 15, 2026

Four Months and Two-Thirds Off: the Open-Weight Gap, Measured Three Ways

Mozilla’s State of Open Source AI report puts the gap between US closed frontier models and the best open-weight Chinese models at 4.4 months. On METR’s task time-horizon method the best closed model finishes a job 1.7 times as long as the best open one can reliably complete — roughly twelve hours against seven, on Mozilla’s CTO’s illustration. Eight of the ten most-used models on OpenRouter by token volume in August published open weights. And on Terminal-Bench 2.1, run by Vals AI with every model on the same harness, the open-weight GLM 5.2 scored within a point of Claude Opus 4.7 and 4.8 at about a fifth of the cost per completed task. Mozilla argues for open models, which is a stake in its own finding.

Read

Ars Technica

Aug 31, 2026

Three Points Now Separate the Best Model From the Best One You Can Download

Artificial Analysis’ index today puts Claude Opus 5 at 63, the highest of the 128 reasoning models it scores, and Moonshot’s Kimi K3 at 60, the highest of the open-weights models. Behind Kimi come Alibaba’s Qwen3.8 at 58 and DeepSeek V4 Pro at 53. Of the 161 models the index evaluates, 86 — more than half — can be downloaded. The index version was also checked this cycle: every row added to this page’s board since June is scored on Intelligence Index v4.1.1, which is what the live index runs today, so the board compares like with like. Nothing needed correcting.

Read

Artificial Analysis

Aug 28, 2026

Tencent Releases an Open-Weight 770-Billion-Parameter Model

Tencent published Hy4 preview, a downloadable mixture-of-experts model with 770 billion parameters — about 49 billion active for any request — and a context window over a million tokens, also served through Tencent Cloud and OpenRouter at $0.83 per million input tokens and $2.50 output. It adds another Chinese open-weight entrant behind Moonshot, Zhipu, Alibaba and DeepSeek. Tencent flagged that the early-release model can over-verify its answers and run long on hard questions. It carries no independent capability score yet — Artificial Analysis has not evaluated it — so it stays off this page's ranked board until one exists.

Read

ReutersTencent

Aug 28, 2026

Hy4 preview ranks among the top tier of open-source models; in a Tencent-internal blind evaluation of 203 engineering tasks by 163 experts it averaged 2.99 out of 4.00, slightly ahead of GLM-5.3 (2.92) and Kimi K3 (2.94).

— Tencent

Read

TencentArtificial Analysis

Aug 26, 2026

A Model You Can Download Ties GPT-5.5, at a Thirty-Third of the Price

Alibaba released Qwen3.8-Flash-Next under open weights, and Artificial Analysis scored it 56 on its Intelligence Index — the same as OpenAI's GPT-5.5 and xAI's Grok 4.5. GPT-5.5 costs $5 per million input tokens; this costs $0.15, with output at $0.47 against $30. It is a mixture-of-experts model with a 125-billion-parameter main body, 51 billion more in N-gram embedding tables, and six billion parameters active per token, running 262,144 tokens of context natively and a million with YaRN. Alibaba presents it as an early look at the architecture Qwen4 will use, and says training cost about a ninth of Qwen3.7-Plus — its own figure, unverified. The independent part is the index score, which is why it is on the board; the lab's own benchmark table is not used.

Read

Artificial AnalysisAlibaba (Qwen team)Hugging Face

Aug 21, 2026

Open Models Are Catching Up Twice as Fast Each Generation

SemiAnalysis measured the open-versus-closed gap across three eras of language models — early scaling, reasoning, agentic — and found it moves in cycles rather than one direction. A frontier lab jumps ahead at the start of an era; others reverse-engineer the advance and close it. What has changed is the clock: with each generation, open-source models take half as long to catch up to the first closed model of that era. The method matters. Rather than score every model on one benchmark set, which saturates and is abandoned, SemiAnalysis scored each era on its own benchmarks and compared catch-up time — a rate, not a level. It assesses GLM 5.3 and Kimi K3 as genuinely able to do the coding and agentic work that carried Anthropic past $65 billion in annualised revenue.

Read

SemiAnalysis

Our reading

Aug 17, 2026

The weights Alibaba published are not the model Alibaba sells

Alibaba's model card for the downloadable Qwen3.8-2.4T-A95B states that the hosted Qwen3.8-Max "is the official version based on Qwen3.8-2.4T-A95B with more features, such as vision input & non-thinking support, 1M context length by default, official built-in tools", and separately that the downloadable model "is a text-only model that requires thinking mode for all interactions. Multimodal inputs are not supported, and thinking cannot be disabled." Alibaba's own page for the hosted model lists the thinking toggle and a 983,000-token maximum input. So the largest set of weights anyone can download is a text-only, always-reasoning sibling of the product on sale, and the score of 58 that places Qwen3.8-Max second among open-weight models on the Artificial Analysis index was measured on the hosted version rather than on the published weights. The smaller Qwen3.8-27B, released two days after the flagship weights, does take image input.

Compared

Hugging FaceQwen Cloud (Alibaba)Artificial Analysis

Aug 17, 2026

Qwen3.8-27B has "excellent capabilities" in coding, professional work, research and long-horizon agentic tasks, and matches the performance of a model ten times its size.

— Alibaba (Qwen team) Mixed

Read

CNBCArtificial Analysis

Aug 14, 2026

A model that runs on a laptop scores level with one ten times its size

Alibaba published Qwen3.8-27B, a dense 27-billion-parameter model under the permissive Apache 2.0 licence, and on the independent Artificial Analysis Intelligence Index it scores 52 — the same score as DeepSeek's V4-Flash, a 284-billion-parameter mixture-of-experts model ten times its size, and level with OpenAI's GPT-5.6 Luna. It is seventeen points above Meta's Muse Glimmer, the 30-billion-parameter open model released four days earlier into the same size class, and eleven points below the best model anyone can buy. The unquantised weights are 55.6 GB; a community four-bit conversion for Apple silicon is 16.1 GB, small enough for a 32 GB Mac with room left for working context. The model takes text and images and carries a 262,144-token context window, extensible to a million. The same index records what the score costs: it generated 160 million tokens to finish the test battery against a 43 million median for open models of its size, reaching its answers by thinking several times longer than its peers. Alibaba's own benchmark comparisons, including a claimed parity with Claude Opus 4.6 at maximum effort and a sweep of its own chosen tests against Muse Glimmer, are the lab's claims and not the independent measure.

Read

Artificial AnalysisThe New StackHugging FaceHugging FaceCNBC

Aug 12, 2026

Alibaba publishes Qwen3.8-Max weights, the largest downloadable model

Alibaba published the weights of Qwen3.8-Max, its 2.4-trillion-parameter flagship (95 billion active per token, one-million-token context) — the first Max-class Qwen released as open weights and the largest-parameter model anyone can download, fulfilling the release the company had promised at the model's 3 August launch. The terms let most developers use, copy, modify and distribute the model without payment, but require a separate paid commercial licence for any 'model as a service' or 'AI work assistant' business whose aggregate revenue exceeds US$50 million over any consecutive 12-month period; internal use is exempt. The model scores 58 on the Artificial Analysis Intelligence Index v4.1.1, second among open-weight models behind Moonshot's Kimi K3 (60) and three points below the closed leader, Claude Opus 5 (63). Artificial Analysis's own model page still classified Qwen3.8-Max as proprietary as of 13 August, a lag behind the actual release. A paid API remains available at $2 per million input tokens and $6 per million output.

Read

South China Morning PostArtificial Analysis

Aug 10, 2026

Meta will 'soon' release open weights for its flagship Muse Spark 1.2, extending its open-weight pivot from the small Muse Glimmer to a frontier-band model.

— Meta AI Mixed

Read

CNBCVentureBeatArtificial AnalysisArtificial AnalysisHugging FaceArtificial Analysis

Aug 10, 2026

Meta returns to open weights with Muse Glimmer

Meta released Muse Glimmer, a 30-billion-parameter open-weight model under the permissive Apache 2.0 licence with weights on Hugging Face — its first open-weight release since retooling its Muse flagship line as proprietary under Alexandr Wang. Optimized for always-on local agents, it is small enough to run on a Mac or PC with a single consumer GPU: compressed to roughly 4-bit precision it shrinks to under 20 GB, and Meta says it is distilled from the larger Muse Spark. It is multimodal (text and image), spans a 131,000-token context and was trained to a January 2026 knowledge cutoff. On the independent Artificial Analysis Intelligence Index v4.1.1 it scores 35 — an edge-scale rather than frontier score, but well above the median for open-weight models of its size. Meta paired the release with a pledge to open the weights of its frontier-band flagship Muse Spark 1.2 (57 on the same index), reopening the open-versus-closed question the company had appeared to abandon. Meta's own benchmark comparisons — against Google's Gemma 4 31B and Alibaba's Qwen3.6-27B — are the lab's own claims; the score cited here is the independent index's.

Read

Meta AI (research blog)Hugging FaceArtificial AnalysisVentureBeatCNBC

Jul 31, 2026

DeepSeek releases V4-Flash, the cheapest well-known model to run

DeepSeek released V4-Flash (V4-Flash-0731), an open-weights reasoning model published under the MIT licence with weights on Hugging Face and designed to run on Huawei Ascend accelerators as well as Nvidia. A 284-billion-parameter mixture-of-experts model activating 13 billion parameters per token, it scores 50 on the Artificial Analysis Intelligence Index v4.1 — level with Google's Gemini 3.6 Flash and a point below Zhipu's GLM-5.2 and Meta's Muse Spark 1.1 — at a published price of $0.14 per million input tokens and $0.28 per million output. Artificial Analysis found it the cheapest well-known model to run, at roughly 3 cents to complete its Intelligence Index test battery, against 86 cents for Moonshot's Kimi K3, $1.86 for OpenAI's GPT-5.6 Sol and $3.15 for Anthropic's Claude Fable 5 — a cost measured per test rather than per token, because a model that is cheap per token can still run up a large bill when it needs more steps to reach an answer.

Read

Artificial AnalysisQuartzThe Next Web

Our reading

Jul 30, 2026

The index moves the largest open-weight model onto the open side of its own cut

Artificial Analysis now classifies Kimi K3 as an open-weights model. Its page for the model carries the label "Open weights model" and answers the question directly — "Yes, Kimi K3 is open weights. The model weights are publicly available and can be downloaded for self-hosting" — listing the weights against the Hugging Face repository that hosts them and the bespoke Kimi K3 License. Before the weights were published on 27 July the same page answered the opposite, that the model was proprietary and its weights not publicly available, which left the index's separate open-weights-against-proprietary ranking counting the largest open-weight model ever released on the wrong side of the distance it exists to measure. The score is unchanged at 57, so what has moved is the cut and not the measurement: the four-point gap to Claude Opus 5 at the top of the index is now readable from Artificial Analysis's open-weights view rather than only by comparing its per-model pages. One difference remains between the index and the licence it describes — Artificial Analysis summarises Kimi K3 as requiring a separate agreement for commercial use, where the licence text grants rights to use, modify, distribute and sell outright and imposes a separate agreement only on operators reselling model access above US$20M in group revenue.

Compared

Artificial AnalysisHugging FaceHugging Face

Earlier this year

Jul 27, 2026

Moonshot publishes the Kimi K3 weights, the largest open-weight model yet released

Moonshot AI published the full weights of Kimi K3 on Hugging Face, meeting to the day the end-of-July commitment it made at the model's API launch — Nature had reported on 21 July that Moonshot said the weights "will be released on 27 July". K3 is a 2.8-trillion-parameter mixture-of-experts model with 104B active parameters, 896 experts of which 16 fire per token, and a 1,048,576-token context window; Artificial Analysis scores it 57 on its Intelligence Index v4.1, unchanged from the API-only release, at $3.00/$15.00 per million tokens. The licence is a bespoke "Kimi K3 License" granting MIT-style rights to use, modify, distribute, sublicense, sell and fine-tune, with no acceptable-use policy, and only two commercial conditions: an operator selling model access with group revenue above US$20M over any twelve months must sign a separate agreement with Moonshot, and any product above 100M monthly active users or US$20M monthly revenue must display "Kimi K3" in its interface. Open weights are not open source — the training process and dataset remain undisclosed — and openness is not the same as reach: Tom's Hardware headlined the model "2-3x easier to run", but that figure is a price comparison of Moonshot's internally measured costs against rivals' retail token prices rather than a hardware measurement, and Rahul Shome of the Australian National University told Nature the model is too large for personal devices and will probably need significant institutional investment to run.

Read

Hugging FaceHugging FaceTom's HardwareArtificial AnalysisNature

Jul 27, 2026The open-weight gap on the Artificial Analysis index halves in four days

Jul 26, 2026Inside the open-weight tier, prices span more than an order of magnitude

Jul 17, 2026“Moonshot AI claims Kimi K3 beat Claude Opus 4.8 and GPT 5.5 — models CNBC describes as sitting just behind Anthropic's and OpenAI's leading-edge systems — on benchmarks including coding and general agents.” — Moonshot AI^{±Fact check: mixed}

Jul 16, 2026

Moonshot's Kimi K3 puts a Chinese model in the frontier's top tier

At 2.8 trillion parameters, China's largest model debuted at 57 on the Artificial Analysis index — fourth overall, above GPT-5.5 and behind only Claude Fable 5 and GPT-5.6 Sol — and second on AA's agentic knowledge-work benchmark. Chinese AI rivals' shares plunged on the news (Z.ai -28%, MiniMax -16%). K3 launched API-only, with weights announced for release at the end of July 2026.

Read

Moonshot AICNBCArtificial AnalysisUK AI Security InstituteArtificial Analysis

Jul 15, 2026Thinking Machines releases Inkling, a US open-weight frontier entrant

Jul 9, 2026

Meta ships its flagship model closed

Meta released Muse Spark 1.1, a multimodal reasoning model with a 1,048,576-token context window, alongside the public preview of the Meta Model API — its first developer API for the model. Artificial Analysis scores it 51 on its Intelligence Index v4.1 at the highest effort setting, against a median of 33 for reasoning models in the same price band, and prices it at $1.25 per million input tokens and $4.25 per million output tokens, undercutting Zhipu's GLM-5.2 at the identical score of 51. It is also closed: Artificial Analysis lists the model as proprietary and states its weights are not publicly available, Meta discloses no parameter count, and Meta's own launch post never uses the words "open weights" or "open source" about it. That is a reversal for the company whose Llama 2 release in July 2023 turned open-weight frontier-chasing from a leak into a corporate strategy, on the argument that "by making AI models available openly, they can benefit everyone". Artificial Analysis measures the model as fast — 129 output tokens per second against a 67 median — and somewhat verbose, consuming 94 million output tokens to run the index against a 63 million median.

Read

Meta AIArtificial AnalysisMeta AIArtificial Analysis

Jul 1, 2026UK AISI: the open-weight cyber gap has narrowed to 4-7 months

Jun 16, 2026Zhipu's GLM-5.2 becomes the strongest open-weight model ever released

May 31, 2026MiniMax M3 brings 1M-token context to open-weight pricing

Apr 24, 2026DeepSeek previews V4, open-source in Pro and Flash versions

Feb 17, 2026“The most likely (by far) outcome is for the status quo to continue and for the best open models to lag the best closed models by 6-9months.” — Nathan Lambert

Dec 1, 2025DeepSeek's V3.2 claims GPT-5-level daily-driver performance

Earlier

Aug 5, 2025OpenAI returns to open weights with gpt-oss
Apr 28, 2025Alibaba's Qwen3 brings hybrid reasoning to open weights at every scale
Apr 7, 2025“Llama is no longer the open standard.” — Nathan Lambert✓Fact check: true
Jan 20, 2025
DeepSeek releases R1, an open-weight reasoning model priced at ~2% of o1R1 shipped under an MIT license with API pricing of $0.55 per million input tokens and $2.19 output — versus $15/$60 for the o1 it claimed parity with on math, code, and reasoning benchmarks — alongside six distilled smaller models. The release triggered a global market repricing of AI economics within a week and made a Chinese lab, for the first time, the reference point in the frontier conversation.
Read
DeepSeekapi-docs.deepseek.com50% credibilityEarned from 1 checked claim.CNBCcnbc.com80% credibility · not yet testedAn opening estimate from its institutional history — none of its claims checked yet.
Dec 26, 2024DeepSeek-V3: frontier-adjacent open weights at a fraction of the training cost
Jul 23, 2024Meta's Llama 3.1 405B is the first open-weights model in the frontier class
Dec 11, 2023Mistral releases Mixtral 8x7B, open weights under Apache 2.0
Jul 18, 2023Meta releases Llama 2 with open weights for commercial use
Jul 18, 2023“We believe an open approach is the right one for the development of today's AI models, especially those in the generative space where the technology is rapidly advancing. By making AI models available openly, they can benefit everyone.” — Meta AI
Mar 3, 2023Meta's LLaMA weights leak onto the open internet

Covered since Mar 2023Read it from the start →