https://shipped.anth.us/Announcements and real-world deployments of AI, checked against their sources and scored: open-weights releases, decision models in the wild, and what is actually in use.enWed, 21 Oct 2026 00:00:00 +0000https://shipped.anth.us/decisions/decisionmodels-reax-1-1/https://shipped.anth.us/decisions/decisionmodels-reax-1-1/Apache-2.0 weights built on REAX 1, itself a fine-tune of pplx-decider-v1.2. The card reports 71.27 on its own SN92 scoring.Wed, 21 Oct 2026 00:00:00 +0000decision-modelsclassifierreaxhttps://shipped.anth.us/decisions/polignano-decision4rec-kev/https://shipped.anth.us/decisions/polignano-decision4rec-kev/Apache-2.0 adapters from 0.8B to 27B that read a pool of 30 candidates and output a probability for each. Development results only.Mon, 19 Oct 2026 00:00:00 +0000decision-modelsrankingclassifiermarco-polignanohttps://shipped.anth.us/decisions/accuknox-guard-1-1/https://shipped.anth.us/decisions/accuknox-guard-1-1/A 4.5B guard fine-tuned from a Qwen3.5-4B base that returns a block probability. The card reports 0.802 on a 27,416-item safety set.Fri, 16 Oct 2026 00:00:00 +0000decision-modelsguard-modelclassifieraccuknox-inchttps://shipped.anth.us/decisions/maincode-matilda-decision-v1/https://shipped.anth.us/decisions/maincode-matilda-decision-v1/Apache-2.0 weights on Qwen3.8-27B. The maker's own run scores 65.38 on the public Jev Decision Index 0.3 suite.Thu, 15 Oct 2026 00:00:00 +0000decision-modelsclassifiermaincodehttps://shipped.anth.us/decisions/decisionmodels-s1-llm-auto-router/https://shipped.anth.us/decisions/decisionmodels-s1-llm-auto-router/A pruned mmBERT-small that answers seven routing questions in one CPU pass. The card reports 0.898 category accuracy on its held-out set.Thu, 15 Oct 2026 00:00:00 +0000decision-modelsrouterclassifierdecisionmodelshttps://shipped.anth.us/open-weights/liquidai-lfm2-5-24b-a2b/https://shipped.anth.us/open-weights/liquidai-lfm2-5-24b-a2b/LFM2.5-24B-A2B is a 24B-parameter reasoning model with about 2.45B active, built for agents on-device, under the LFM1.0 licence.Mon, 12 Oct 2026 00:00:00 +0000open-llmsmixture-of-expertssmall-modelmultilingualliquid-aihttps://shipped.anth.us/decisions/startlux-decision-v1-1/https://shipped.anth.us/decisions/startlux-decision-v1-1/StartLux-Decision-v1.1 ships Instant and AutoThink variants. Weights are CC BY-NC 4.0; commercial use needs a separate licence.Sun, 11 Oct 2026 00:00:00 +0000decision-modelsclassifierstartluxhttps://shipped.anth.us/decisions/aisingapore-sea-lion-guard-v2/https://shipped.anth.us/decisions/aisingapore-sea-lion-guard-v2/A 31B decision model (plus E4B and E2B) for content moderation and action governance across eight languages. Scores AUPRC 93.5 on general safety.Sun, 11 Oct 2026 00:00:00 +0000decision-modelsguard-modelpublic-sectorai-singaporehttps://shipped.anth.us/decisions/microsoft-decision-1/https://shipped.anth.us/decisions/microsoft-decision-1/Microsoft-Decision-1 returns a probability per answer option instead of prose. Hosted on Foundry and OpenRouter; no licence stated.Fri, 09 Oct 2026 00:00:00 +0000decision-modelsclassifierroutermicrosofthttps://shipped.anth.us/decisions/amazon-strands-decider-2b/https://shipped.anth.us/decisions/amazon-strands-decider-2b/A 2B LoRA on Qwen3.5-2B-Base, Apache-2.0, with a /v1/systemone server. The card reports 0.779 on 231 public JevBench tasks.Fri, 09 Oct 2026 00:00:00 +0000decision-modelsclassifierrouteramazonhttps://shipped.anth.us/open-weights/reflection-beam-1-0-501b-a23b/https://shipped.anth.us/open-weights/reflection-beam-1-0-501b-a23b/Beam is Reflection's first open model: 501B total and 23B active parameters, 1M-token context, under Apache 2.0.Wed, 07 Oct 2026 00:00:00 +0000open-llmsmixture-of-expertsreasoning-modelcode-modelreflectionhttps://shipped.anth.us/decisions/openai-decisions-api/https://shipped.anth.us/decisions/openai-decisions-api/A /v1/decisions endpoint returns typed predicate, choice and score answers. Public beta, $0.10 per million input tokens, GA promised in weeks.Tue, 06 Oct 2026 00:00:00 +0000decision-modelsrouterclassifieropenaihttps://shipped.anth.us/decisions/torchcast-decision-27b/https://shipped.anth.us/decisions/torchcast-decision-27b/A LoRA fine-tune of StartLux-Decision-27B that speaks the /v1/systemone API. Weights are CC BY-NC 4.0; card reports 65.00 on Index 0.2.1.Sat, 03 Oct 2026 00:00:00 +0000decision-modelsclassifiertorchcast-aihttps://shipped.anth.us/open-weights/aleph-alpha-kolibri-1/https://shipped.anth.us/open-weights/aleph-alpha-kolibri-1/Kolibri 1 is a 78B mixture-of-experts reasoning model with 3.46B active parameters, tuned for German and English, under Apache 2.0.Sat, 03 Oct 2026 00:00:00 +0000open-llmsmixture-of-expertsreasoning-modelmultilingualaleph-alphahttps://shipped.anth.us/decisions/sberai-frida-decisions/https://shipped.anth.us/decisions/sberai-frida-decisions/A 0.8B MIT-licensed encoder that picks, rates, answers yes/no or ranks in one pass. The card reports 0.893 on 735 razvilka items.Fri, 02 Oct 2026 00:00:00 +0000decision-modelsclassifiersberaihttps://shipped.anth.us/decisions/h2oai-h2o-lightning-4b/https://shipped.anth.us/decisions/h2oai-h2o-lightning-4b/Built on Qwen3.5-4B, with GGUF files from 1.43 GB for llama.cpp. The card reports 88.74% on 231 public JevBench text questions.Fri, 02 Oct 2026 00:00:00 +0000decision-modelsclassifierh2o-aihttps://shipped.anth.us/decisions/perplexity-pplx-decider/https://shipped.anth.us/decisions/perplexity-pplx-decider/pplx-decider-v1 and v1.1 are open-weight 27B models that return probabilities for fixed answers. Built on Qwen3.8-27B.Thu, 01 Oct 2026 00:00:00 +0000decision-modelsclassifierrouterperplexityhttps://shipped.anth.us/open-weights/fish-audio-drama-3/https://shipped.anth.us/open-weights/fish-audio-drama-3/Drama 3 is a 35B MoE speech model (3B active) with up to 16 speaker tags and voice cloning; research licence.Thu, 01 Oct 2026 00:00:00 +0000open-ttsmulti-speakervoice-cloningmultilingual-ttsfish-audiohttps://shipped.anth.us/decisions/vllm-decision-2-0/https://shipped.anth.us/decisions/vllm-decision-2-0/Decision 2.0 is an Apache-2.0 family from vllm-sr. The 27B Vega card reports 74.0 on JevArena. Lux is a 9B sibling.Tue, 29 Sep 2026 00:00:00 +0000decision-modelsrouterclassifiervllm-semantic-routerhttps://shipped.anth.us/decisions/chaoliangunsw-jev-style-2b-decision-v3/https://shipped.anth.us/decisions/chaoliangunsw-jev-style-2b-decision-v3/Apache-2.0 fine-tune of Qwen3.5-2B. The card reports 73.6% on public JevBench items against Jev 1.13 at 86.6%.Sun, 27 Sep 2026 00:00:00 +0000decision-modelsclassifierchaoliang-yanhttps://shipped.anth.us/decisions/internlm-intern-decision/https://shipped.anth.us/decisions/internlm-intern-decision/Intern-Decision-4B is fine-tuned from Qwen3.5-4B and answers up to 16 questions per pass, with image input. Apache-2.0.Sat, 26 Sep 2026 00:00:00 +0000decision-modelsclassifiershanghai-ai-laboratoryhttps://shipped.anth.us/open-weights/xiaomimimo-mimo-v2-6-pro-rl/https://shipped.anth.us/open-weights/xiaomimimo-mimo-v2-6-pro-rl/MiMo-V2.6-Pro-RL is the flagship of the MiMo-V2.6 series: 1.02T total, 42B active, 1M context, text, image, video and audio in.Mon, 21 Sep 2026 00:00:00 +0000open-llmsmixture-of-expertsmultimodalreasoning-modelxiaomi-mimohttps://shipped.anth.us/decisions/distil-labs-qwen3-5-4b-invoice-decision/https://shipped.anth.us/decisions/distil-labs-qwen3-5-4b-invoice-decision/A 4B Qwen3.5 fine-tune returning one of five decisions. The card reports 98 of 100 correct on synthetic test invoices.Mon, 21 Sep 2026 00:00:00 +0000decision-modelsclassifierdistil-labshttps://shipped.anth.us/open-weights/jetbrains-mellum2-1-12b-a2-5b-thinking/https://shipped.anth.us/open-weights/jetbrains-mellum2-1-12b-a2-5b-thinking/Mellum2.1 Thinking is a 12B mixture-of-experts reasoning model with 2.5B active, retrained mostly with reinforcement learning on real repositories.Sun, 20 Sep 2026 00:00:00 +0000open-llmsmixture-of-expertscode-modelreasoning-modeljetbrainshttps://shipped.anth.us/decisions/jaredpalmer-kev-9b/https://shipped.anth.us/decisions/jaredpalmer-kev-9b/An independent adapter that implements TypeSafe's System One API. The card reports 0.852 on a locked out-of-domain test.Sun, 20 Sep 2026 00:00:00 +0000decision-modelsclassifierjared-palmerhttps://shipped.anth.us/decisions/convai-laya-typed-decisions/https://shipped.anth.us/decisions/convai-laya-typed-decisions/An Apache-2.0 encoder fine-tuned on four synthetic workflows. The card reports 0.766 accuracy and warns its probabilities are over-confident.Fri, 18 Sep 2026 00:00:00 +0000decision-modelsclassifierconvai-innovationshttps://shipped.anth.us/decisions/typesafe-jev/https://shipped.anth.us/decisions/typesafe-jev/Jev takes state and typed questions and returns probabilities. Early access via API, $0.042 per million input tokens; weights not stated.Tue, 15 Sep 2026 00:00:00 +0000decision-modelsclassifierroutertypesafe-aihttps://shipped.anth.us/open-weights/yandex-aliceai-foundation-80b-a3b-base/https://shipped.anth.us/open-weights/yandex-aliceai-foundation-80b-a3b-base/AliceAI-Foundation-80B-A3B-Base is an 80B base model with 3B active parameters and 262,144-token context, trained from scratch.Sat, 12 Sep 2026 00:00:00 +0000open-llmsmixture-of-expertsbase-modelmultilingualyandexhttps://shipped.anth.us/open-weights/deepseek-ai-deepseek-v4-1-flash/https://shipped.anth.us/open-weights/deepseek-ai-deepseek-v4-1-flash/DeepSeek-V4.1-Flash is a 552B-backbone multimodal MoE with 1M-token context and a 890-byte-per-token KV cache, under MIT.Thu, 10 Sep 2026 00:00:00 +0000open-llmsmixture-of-expertsmultimodalreasoning-modeldeepseek-aihttps://shipped.anth.us/open-weights/openbmb-minicpm5-2b/https://shipped.anth.us/open-weights/openbmb-minicpm5-2b/MiniCPM5-2B is a dense 2B on-device model for coding, tool use and long context, released with its UltraData training sets.Sun, 06 Sep 2026 00:00:00 +0000open-llmssmall-modeldense-modelopenbmb