#01
moonshotai/Kimi-K3
Official Hugging Face model artifact; the card can separate its multimodal capability from the custom license, extreme hardware needs, and not-hands-on-tested status.
Source-checked trend review
50 developer AI signals, sorted into tools, news, and context
A supplied research export ranked 50 names that were moving across developer channels. RepoRadar checked every identity, resolved aliases, separated actual tools from announcements and family labels, and chose a public surface for each one.
The important correction
The supplied ranking mixed Hacker News points, weekly GitHub growth, Hugging Face engagement, launch velocity, and search-style proxies. It did not contain reproducible Google search-volume data, so RepoRadar publishes it as a signal review rather than repeating an unverified 'most searched' claim.
Reviewed: 2026-08-01. Rankings preserve the source order; RepoRadar placement does not endorse the source score.
37Radar cards
5News items
8Guide decisions
How the placement works
Radar card means a concrete repository, model, application, or service that stays useful after the launch cycle. News means a dated release or announcement. Guide means the searched name is a family, protocol, adjacent project, vanished source, narrow skill, or high-risk utility that needs context instead of a normal recommendation.
"Added" means this review created canonical coverage. "Already on radar" points to the existing record. A safety hold is still documented here, but it is not promoted into the normal catalog.
All 50 decisions
#02
google/gemma-4-26B-A4B-it
A specific official model checkpoint is more useful as a comparable card than as a one-day announcement.
#03
block/buzz
Runnable open-source workspace with clear collaboration and agent workflows; permissions and message access belong on the card.
#04
mattpocock/skills
A practical reusable skill library already has canonical RepoRadar coverage; the ranking should link to it rather than duplicate it.
#05
deepseek-ai/DeepSeek-V4-Pro
The official gated checkpoint is a durable model artifact; access, license terms, and infrastructure requirements matter more than launch-day chronology.
#06
diegosouzapw/OmniRoute
The repository already has a canonical routing card under the current owner, so this signal resolves to that record rather than a stale alias.
#07
Gemini Robotics 2
This is an official model-family launch announcement with preview and early-access availability, so it belongs in the chronological News feed rather than the reusable-tool radar.
#08
openai/codex-security
OpenAI publishes a concrete Apache-licensed CLI and TypeScript SDK with installable code, documentation, and a bounded security-scanning workflow, so it warrants a radar card.
#09
OpenAI GPT-5.6
This is a dated model release with existing RepoRadar news coverage and a separate practical Codex setup guide.
#10
koala73/worldmonitor
The existing canonical card covers the deployable situational-awareness app; no alias duplicate is needed.
#11
Claude 5 family (Opus 5 anchor)
The ranked family signal is anchored to Anthropic’s official Opus 5 launch. A model announcement belongs in News; the guide keeps the family-level context without inventing a downloadable card.
#12
Qwen3 family
Qwen3 spans many checkpoints, sizes, modalities, and deployment profiles. A family guide is more useful than one card that hides those differences.
#13
FLUX.1-dev / FLUX.2-dev
The ranking bundles separate model generations and licenses. Keep the comparison at guide level rather than merging them into one misleading radar record.
#14
alibaba/open-code-review
RepoRadar already has a canonical source-linked card for the repository.
#15
deepseek-ai/DeepSeek-R1
A mature, heavily used official reasoning checkpoint deserves evergreen model coverage rather than being treated as fresh news.
#16
citrolabs/ego-lite
It stores logged-in browser state and account data. The guide records the signal and risk, but RepoRadar does not promote it as a normal recommendation without a bounded hands-on review.
#17
ayghri/i-have-adhd
This is a focused Claude Code formatting skill rather than a standalone AI engine. It is useful context for skill design, but too narrow for a separate radar card.
#18
earendil-works/pi
The canonical card already covers the modular agent toolkit, including the shell and file permissions that depend on how its coding-agent tools are enabled.
#19
virgiliojr94/book-to-skill
Existing coverage already explains the document-to-skill workflow and file-processing caveats.
#20
openai/gpt-oss-120b
A specific official open-weight checkpoint is best handled as a model card with realistic hardware and license context.
#21
moonshotai/Kimi-Linear-48B-A3B-Instruct
The vague search term resolves to a specific official MIT-licensed checkpoint and paper-backed architecture, so a canonical model card is warranted.
#22
hexgrad/Kokoro-82M
A compact Apache-licensed text-to-speech model with substantial real use is a durable discovery item; synthetic-audio misuse still needs a caution label.
#23
Scriptc / Vercel scripts
The supplied Vercel path now resolves to a sign-in gate, so its public artifact and claims could not be independently reviewed. The guide preserves that lookup outcome without inventing a card.
#24
1jehuang/jcode
The coding-agent repository already has a canonical RepoRadar card.
#25
stabilityai/stable-diffusion-xl-base-1.0
SDXL remains a concrete, widely deployed base model; a card can surface its custom OpenRAIL license and image-generation caveats.
#26
CoreBunch/Instatic
The self-hostable AI CMS has a real repository, documented stack, and clear deployment workflow; infrastructure and database access require explicit risk context.
#27
moeru-ai/airi
The actively developed local companion has a runnable repository and broad integrations; microphone, game control, and external actions make risk part of the recommendation.
#28
owensong/Inflect-Micro-v2
A specific Apache-licensed low-latency TTS checkpoint is comparable as a model card, not a transient launch headline.
#29
opengeos/GeoLibre
GeoLibre is an interesting geospatial application, but AI is not the core verified function. Keep it as adjacent context rather than dilute the AI radar.
#30
openai/whisper-large-v3
A concrete official speech-recognition checkpoint with strong adoption belongs in the evergreen model catalog.
#31
pascalorg/editor
The editor is a legitimate browser CAD project but has no strong first-party AI function. The guide records the adjacent signal without mislabeling it as AI.
#32
earthtojake/text-to-cad
RepoRadar already covers this exact text-to-CAD repository.
#33
black-forest-labs/FLUX.1-schnell
This is a distinct Apache-licensed fast image model, so it should not be folded into the broader FLUX family note.
#34
Tongyi-MAI/Z-Image-Turbo
The official Apache-licensed checkpoint has strong adoption and a clear fast-generation use case; a model card can state that RepoRadar has not hands-on tested it.
#35
Model Context Protocol specification
The dated protocol revision is an ecosystem update with breaking transport, authorization, and SDK changes, so it belongs in News while this guide records the migration context.
#36
feyninc/FeyNoBg
A downloadable background-removal model with an official launch write-up has a clear bounded use case; the card can flag its custom license.
#37
schildep/verified-3d-mesh-intersection
The repository combines 3D reconstruction with a formally verified CSG predicate and publishes reproducible code under MIT.
#38
pingdotgg/t3code
A runnable local web UI for orchestrating coding agents is a practical workflow tool, with shell and repository modification risk made explicit.
#39
VladUZH/qwen-scribe
A focused speech-transcription package with documented model behavior and MIT code is useful enough for a card; recordings may contain sensitive data.
#40
microsoft/VibeVoice
The existing canonical card covers the Microsoft research toolkit and synthetic-voice risk.
#41
black-forest-labs/FLUX.1-Kontext-dev
This specific image-editing checkpoint has a distinct use case and license, so it benefits from its own card instead of a family umbrella.
#42
Tokenless
The live service provides cache-based model routing with documented provider support; prompts may cross service boundaries, so privacy and routing claims need caution.
#43
SerhiiKorniienko/bullshit-detector
The MIT-licensed skill pack encodes concrete source-verification and confidence-reporting workflows; claims still require checking the primary evidence.
#44
Lightricks/LTX-2.3
A specific video-generation checkpoint with substantial adoption belongs in the model catalog, with its custom license and synthetic-media risk visible.
#45
hamzarehmandeveloper/claude-account
The utility manages Claude credentials or account state. RepoRadar records it here but does not promote it as a normal card without a credential-handling and threat-model review.
#46
funador/claude-code-merge-queue
The repository already has canonical coverage; the ranking links to the existing record.
#47
Qwen/Qwen-Image-Edit-2511
A concrete Apache-licensed image-editing checkpoint is best compared as a model card, with provenance and editing misuse treated as real risks.
#48
Supapool
The live CLI provisions databases, generates code, and can apply migrations; that is practical developer infrastructure with high external-system impact.
#49
Echologue
The local-first voice journal has a clear user workflow and privacy documentation; microphone input and sensitive personal notes still justify a data-risk label.
#50
Tines 3B
This is a dated first-party product announcement, not an evergreen repository or downloadable model artifact, so it belongs in the news feed.
What to do with this list
Start with the cards that match a real workflow, then read the risk and setup notes before installing anything. Treat family-level searches as a prompt to compare exact checkpoints. For session, credential, shell, database, or autonomous-action tools, use a disposable environment and bounded permissions first.
See RepoRadar's methodology for scoring and editorial separation, or browse the full radar.