SoTA Feed — Every open-weights release from the labs that matter

The criterion

Rule: a lab is tracked iff it has at least one model — open or closed weights — scoring strictly above Qwen3.6 27B = 46 on the Artificial Analysis Intelligence Index (v4); what gets blogged is its open conversational releases on Hugging Face. Snapshot: 2026-06-10 · verify at https://artificialanalysis.ai/leaderboards/models. Every row cites its evidence.

Tracked labs (12)

HF orgLabQualifying modelScoreSource
deepseek-aiDeepSeekDeepSeek V4 Pro52artificialanalysis.ai article, 2026-04-24 (V4 Flash: 47)
moonshotaiMoonshot AIKimi K2.6 (Reasoning)54artificialanalysis.ai, 2026-04-30
XiaomiMiMoXiaomi MiMoMiMo-V2.5-Pro (Reasoning)54artificialanalysis.ai, 2026-04-30
zai-orgZ.aiGLM-5.1≈52 (top open model on GDPval-AA: 1535)artificialanalysis.ai DeepSeek-V4 article, 2026-04-24
MiniMaxAIMiniMaxMiniMax-M2.7 / M3≈47–50 (intelligence-vs-size Pareto frontier, ≈ V4 Flash)artificialanalysis.ai, 2026-04
QwenAlibaba QwenQwen3.6 Max (open) / Qwen3.7 Preview≈54 / 57 (baseline Qwen3.6 27B = 46 is theirs)deeplearning.ai The Batch + insiderllm.com, 2026-05
nvidiaNVIDIANemotron 3 Ultra (550B-A55B)47.7artificialanalysis.ai launch article, 2026-06-04
googleGoogleGemini 3.1 Pro Preview (closed)57artificialanalysis.ai, 2026-04; open releases: Gemma 4 line
openaiOpenAIGPT-5.5 xhigh (closed)60artificialanalysis.ai, 2026-04; open releases: gpt-oss line
anthropicAnthropicClaude Opus 4.7 (closed)57artificialanalysis.ai, 2026-04; no open conversational releases yet
xai-orgxAIGrok 4.3 high (closed)53artificialanalysis.ai X post, 2026-04-30; open releases: legacy Grok weights
meta-llamaMetaMuse Spark (closed)52artificialanalysis.ai Muse Spark article, 2026; open releases: Llama line

Watchlist / excluded

HF orgLabStatus
stepfun-aiStepFunStep 3.7 Flash called best open-weight agent; AAII score unverified
inclusionAIAnt LingLing/Ring 2.5 1T-param models; AAII score unverified
tencentTencent HunyuanHunyuan 3.0 295B-A21B; AAII score unverified
ByteDance-SeedByteDance SeedDoubao Seed Code = 34; 'Seed 2.0 Pro matches GPT-5.2' claim unverified on AAII
baiduBaiduERNIE 5.0 (2.4T, closed) — no AAII data found
mistralaiMistralMistral Large 3 scores 23 — far below baseline
amazonAmazonNova line — nothing found above the bar on AAII as of snapshot
microsoftMicrosoftMAI / Phi — nothing found above the bar on AAII as of snapshot

Filter for "conversational release": Hugging Face conversational tag (present iff the repo ships a chat template) and pipeline ∈ {text-generation, image-text-to-text, any-to-any}.