The Open Weights
LatestModelsLeaderboardsCompanies
Subscribe
The Open Weights

The daily record of open-source AI. New model releases, leaderboards, and what's coming next — written for people who ship.

Refreshed every 12 hours

Discover

  • Latest releases
  • New today
  • Trending models

Browse

  • All models
  • Companies
  • Categories
  • Leaderboards

About

  • About
  • Editorial policy
  • Hardware estimates
  • RSS feed
  • llms.txt
  • Newsletter

© 2026 The Open Weights. An independent publication.

PrivacyTermsSMSAggregated by Claude · curated by humans.
LatestInternlmS2.1-Flash
InternlmReasoning

InternLM's Sage S2.1 Flash targets scientific reasoning

The 320B mixture-of-experts model is built to reason over scientific evidence and act inside interactive environments.

Nov 15, 2026
Major releaseOther
Sage S2.1 Flash

InternLM has released Sage S2.1 Flash, a large mixture-of-experts model designed less for general chat and more for a specific job: reasoning through scientific evidence and taking actions inside interactive environments. The model is available on Hugging Face.

At roughly 320 billion total parameters, Sage S2.1 Flash leans on a mixture-of-experts (MoE) design, which activates only a fraction of its weights per token to keep inference costs manageable relative to its raw size. InternLM positions it as a reasoning-first system with multimodal inputs, rather than a single-purpose text model.

Why it matters

Most frontier open-weights releases chase broad assistant benchmarks. Sage's framing is narrower and more interesting:

  • Evidence-grounded reasoning over scientific material, rather than open-ended generation
  • Agentic behavior, meaning the model is meant to act within interactive environments and tools
  • An MoE architecture that trades sparse activation for scale

That combination points toward research and lab-style workflows, where tracing a conclusion back to evidence matters as much as the answer itself.

The model ships under a non-standard license, so teams evaluating it for production should read the terms on the model card carefully. InternLM links its accompanying paper via Hugging Face for those who want the methodology behind the release.

Sources

  • internlm/Sage-S2.1-Flash

    Hugging Face

    Visit
OlderMeituan's LongCat Adds Audio-Driven Avatar VideoMeituan · Image → Video · last weekNewerApple's FastFrames-L Targets Real-Time Video GenerationApple · Text → Video · 4 days ago

Get the model

Hugging Face

Specs

Parameters320B · MoE
Context window1.0M tokens
Size642.7 GB
PrecisionBF16
ArchitectureGlm5NextForConditionalGeneration
LicenseOTHER
Downloads1.1K
Likes186

Can you run it?

Needs a multi-GPU server — about 194 GB at 4-bit.

  • BF16 (as published)

    Multi-node cluster

    644.1 GB
  • 8-bit

    8×H100 80 GB node

    341.6 GB
  • 4-bit

    4×H100 80 GB · 512 GB Mac Studio

    194 GB
Your machine
  • Mixture-of-experts: every expert must be loaded, so memory follows total parameters, not the active slice.
  • KV cache sized for 8,192 tokens of context; longer prompts need proportionally more.

Based on Hugging Face weights metadata. Assumes an 8K context and ~1 GB runtime overhead; actual needs vary. How we estimate


Modalities

Text / LLMAny-to-AnyReasoning

The Weekly Weights

Every open release that mattered, one email a week.

0 comments

No comments yet. Be the first to weigh in.

More from Internlm

All Internlm releases →
Atria Dawn Preview
Internlm/ReasoningDatacenter

InternLM's Atria Dawn Preview Targets Agentic Tasks

A new mixture-of-experts model trained on verified tool interactions arrives as an early preview under an MIT license.

Sep 11, 2026
Intern-S2-Preview-397B
Internlm/Vision-LanguageMulti-GPU server

InternLM Previews 397B Vision-Language Model

The Intern-S2 preview arrives as a very large multimodal system under a permissive Apache-2.0 license.

Jul 16, 2026

More in Reasoning

All Reasoning →
ServiceNow/Text → SpeechConsumer GPU

ServiceNow Debuts Apriel Voice 27B, an Open Speech Model

The 27-billion-parameter preview unifies text-to-speech, transcription, and reasoning in a single open model.

Oct 28, 2026
SenseNova-Loop-9B
SenseTime/Vision-LanguageRuns on a laptop

SenseTime previews a 9B looped-transformer VLM

SenseNova-Loop-9B-Preview puts a recurrent-depth architecture to work on multimodal reasoning.

Oct 28, 2026
DeepSeek-V4.1-Flash
DeepSeek/Text / LLMDatacenter

DeepSeek Releases V4.1-Flash, an Open Multimodal MoE

The new mixture-of-experts model handles text, vision, and reasoning, and ships under a permissive MIT license.

Oct 17, 2026