Dekibyu — Open-model intel
Data as of 2026-09-21
Snapshot↻ Auto-refreshes every 15 min
Radar.
The open-weight models worth your GPU right now — best for code, image and video, plus fresh drops. Refreshed live from Hugging Face on every visit.
18
Models ranked
63,750
Community likes
Qwen3.8-27B
Newest drop · 2026-08-05
00 Fresh drops
Fresh drops
Notable releases from the last 120 days, ranked by community likes.
Qwen · 2026-08-05
Qwen3.8-27B
♥ 15,935 · apache-2.0
moonshotai · 2026-06-13
Kimi-K3
♥ 11,455 · other
MiniMaxAI · 2026-07-28
MiniMax-H3
♥ 5,556 · other
Qwen · 2026-08-24
Qwen3.8-Flash-Next
♥ 5,535 · other
zai-org · 2026-06-16
GLM-5.2
♥ 5,128 · mit
Lightricks · 2026-07-23
LTX-2.5
♥ 4,628 · other
baidu · 2026-06-19
Unlimited-OCR
♥ 4,270 · mit
deepseek-ai · 2026-07-31
DeepSeek-V4-Flash-0731
♥ 3,988 · mit
01 Open-model intel
Best for code
Ranked by community likes on Hugging Face — the closest thing to a live open-model vote.
02 Open-model intel
Best for image
Ranked by community likes on Hugging Face — the closest thing to a live open-model vote.
03 Open-model intel
Best for video
Ranked by community likes on Hugging Face — the closest thing to a live open-model vote.
04 Judged rankings
Arena ELO — judged by humans
Blind human-preference votes from LMArena's official open dataset — proprietary models included, open weights badged.
04a Best for code · LMArena webdev
★ Best open — kimi-k3-max ELO 1674 · #5 overall
- 01ELO 1800
gpt-6-astra-max
openai · Proprietary · 2,281 votes
- 02ELO 1758
claude-fable-5.1-max
anthropic · Proprietary · 3,036 votes
- 03ELO 1687
claude-opus-5-max
anthropic · Proprietary · 12,087 votes
- 04ELO 1681
qwen3.8-max-0902
alibaba · Proprietary · 2,262 votes
- 05ELO 1674
kimi-k3-max Open
moonshot · Kimi K3 license · 4,547 votes
- 06ELO 1671
qwen3.8-max
alibaba · Proprietary · 3,221 votes
04b Best for image · LMArena text_to_image
★ Best open — ideogram-4.0-quality ELO 1204 · #17 overall
- 01ELO 1421
gpt-image-2.5-sunburst
openai · Proprietary · 3,149 votes
- 02ELO 1399
gpt-image-2.5-flare
openai · Proprietary · 2,856 votes
- 03ELO 1381
gpt-image-2 (medium)
openai · Proprietary · 78,731 votes
- 04ELO 1331
mai-image-2.6
microsoft-ai · Proprietary · 11,213 votes
- 05ELO 1315
grok-imagine-image-2.0 (low)
xai · Proprietary · 2,681 votes
- 06ELO 1301
reve-2.1
reve · Proprietary · 7,573 votes
04c Best for video · LMArena text_to_video
★ Best open — minimax-h3 ELO 1462 · #8 overall
- 01ELO 1515
gemini-omni-1.1-flash
google · Proprietary · 1,777 votes
- 02ELO 1511
gemini-omni-flash
google · Proprietary · 21,934 votes
- 03ELO 1494
wan3.0
alibaba · Proprietary · 1,167 votes
- 04ELO 1494
flux-3-video
bfl · Proprietary · 1,290 votes
- 05ELO 1491
grok-imagine-video-1.5-agent
xai · Proprietary · 1,195 votes
- 06ELO 1482
dreamina-seedance-2.5-720p
bytedance · Proprietary · 4,066 votes
05 Judged rankings
Judged rankings
For blind human-preference ELO, cross-check with the LMArena leaderboards.
06 How this reads
How this reads
“Open” here means open weights, not any single licence — check the licence chip before commercial use. GGUF quants are skipped so base models rank. Numbers update on every page load; if the network fails you see the last build-time snapshot.