| payload |
{"created_at":"2026-04-20T22:38:30.834 {"created_at":"2026-04-20T22:38:30.834831+00:00","dedupe_key":"signal_enriched:discovery_crypto_extremes_delta:47bc063643140b37","evidence_event_ids":["evt_1957056c6364"],"signal_type":"discovery_crypto_extremes_delta","source":"discovery_ingestor","value":{"aggregator_url":"https://decrypt.co/364948/alibaba-qwen-3-6-max-preview-most-powerful-model","as_of":"2026-04-20T22:38:30.834831+00:00","canonical_url":"https://decrypt.co/364948/alibaba-qwen-3-6-max-preview-most-powerful-model","enrichment":{"aggregator_url":"https://decrypt.co/364948/alibaba-qwen-3-6-max-preview-most-powerful-model","article_chars":4681,"article_truncated":false,"blocked_reason":null,"candidate_id":"sc_05bf129dc38e5db0","canonical_host":"decrypt.co","canonical_is_aggregator":false,"canonical_url":"https://decrypt.co/364948/alibaba-qwen-3-6-max-preview-most-powerful-model","content_type":"text/html; charset=utf-8","enriched_at":"2026-04-21T21:30:18.435411+00:00","extraction_method":"trafilatura","fetched_description":"Alibaba released Qwen3.6-Max-Preview, its most powerful AI model yet, topping six major coding benchmarks and posting gains in world knowledge and instruction following over its predecessor.","fetched_title":"Alibaba Drops Qwen 3.6 Max Preview\u2014Its Most Powerful Model Yet - Decrypt","final_url":"https://decrypt.co/364948/alibaba-qwen-3-6-max-preview-most-powerful-model","html_truncated":true,"paywall_likely":false,"publisher_domain":"decrypt.co","publisher_resolution":"canonical_url","requested_url":"https://decrypt.co/364948/alibaba-qwen-3-6-max-preview-most-powerful-model","source_event_id":"evt_1957056c6364","source_quality":"high","status_code":200,"version":"signal_enrichment_v2"},"fp":"d490bb303bf77750","kind":"crypto_extremes","published_at":"2026-04-20T21:24:53+00:00","publisher_domain":"decrypt.co","signal_understanding":{"analysis_basis":"article","claim_confidence":0.74,"dates_mentioned":["April 20, 2026","three days after Alibaba open-sourced Qwen 3.6-35B-A3B (relative timing)","last week (Decrypt report timing)","late 2024","end of 2025"],"entities":[{"asset_class":"equity","name":"Alibaba","relevance":"high","symbol":"","type":"company"},{"asset_class":"ai_model","name":"Qwen 3.6-Max-Preview","relevance":"high","symbol":"","type":"model"},{"asset_class":"ai_model","name":"Qwen 3.6-Plus","relevance":"medium","symbol":"","type":"model"},{"asset_class":"software","name":"Qwen Studio","relevance":"medium","symbol":"","type":"platform"},{"asset_class":"software","name":"Alibaba Cloud Model Studio API","relevance":"medium","symbol":"","type":"platform"},{"asset_class":"social_media","name":"Qwen (Alibaba_Qwen)","relevance":"low","symbol":"","type":"account"}],"event_type":"listing","information_gaps":["The article does not provide the exact benchmark scores/rankings\u2014only that the model ranked first across listed benchmarks.","The article does not specify the full methodology or evaluation settings for the cited benchmarks.","\u201cCompatible with OpenAI and Anthropic specifications\u201d is stated, but the exact compatibility details (e.g., endpoints/parameters) are not included.","The article does not provide pricing for the hosted API or the \u201cprice tier\u201d used in the Artificial Analysis comparison.","The article does not include the full list of \u201csix major coding benchmarks\u201d beyond the names provided; it does not clarify whether all are distinct or how they map to \u201csix.\u201d","The article mentions preserve_thinking but does not define its technical behavior beyond carrying reasoning traces across multi-turn conversations."],"key_facts":["Alibaba unveiled Qwen 3.6-Max-Preview as an early preview of its next flagship model.","The model is described as the most powerful Alibaba has shipped so far.","The article states Qwen 3.6-Max-Preview topped six major coding/agentic benchmarks.","Benchmarks listed include SWE-bench Pro, Terminal-Bench 2.0, SkillsBench, QwenClawBench, QwenWebBench, and SciCode.","The article states it posted gains over Qwen 3.6-Plus in world knowledge and instruction following.","Knowledge benchmark gains cited: SuperGPQA +2.3% and QwenChineseBench +5.3%.","Instruction-following benchmark cited: ToolcallFormatIFBench, where it is stated to beat Claude.","The model is available now via Qwen Studio and the Alibaba Cloud Model Studio API under the string qwen3.6-max-preview.","The article states Qwen 3.6-Max-Preview is proprietary/hosted with no open weights.","The article states the API is compatible with OpenAI and Anthropic specifications.","The article states lower-end Qwen models are still open source.","The article says Qwen 3.6-Max-Preview is labeled a work in progress and expects further gains in future versions.","The article states the model supports a 256k token context window and is text-only at launch (no image input).","The article states the model ships with preserve_thinking to carry reasoning traces across multi-turn conversations.","The article states Alibaba recommends preserve_thinking for agentic tasks where context continuity matters.","The article says the release came three days after Alibaba open-sourced Qwen 3.6-35B-A3B (35B parameters, activating 3B per inference).","The article references Decrypt reporting that Alibaba shut down the free tier of Qwen Code shortly after MiniMax rewrote an open-source license to block commercial use without written authorization.","The article claims Qwen overtook Meta\u2019s Llama as the most deployed self-hosted model and attributes momentum to free access.","The article provides a usage trend claim: Chinese open models rose from 1.2% of global open-model usage in late 2024 to roughly 30% by end of 2025, with Qwen leading.","The article cites independent benchmarking from Artificial Analysis placing the model second behind Muse Spark and above the median of comparable reasoning models in its price tier."],"numeric_claims":[{"label":"SuperGPQA increase vs predecessor","value":"+2.3%"},{"label":"QwenChineseBench increase vs predecessor","value":"+5.3%"},{"label":"Qwen 3.6-35B-A3B parameters","value":"35B"},{"label":"Qwen 3.6-35B-A3B activated parameters per inference","value":"3B"},{"label":"Token context window","value":"256k"},{"label":"Chinese open models share late 2024","value":"1.2%"},{"label":"Chinese open models share end of 2025","value":"~30%"}],"primary_claim":"Qwen 3.6-Max-Preview is Alibaba\u2019s most powerful model to date, topping six major coding/agent benchmarks and showing gains in world knowledge and instruction following over Qwen 3.6-Plus.","relevance_score":0.78,"sentiment":"mixed","source_quality":"high","summary":"Alibaba released Qwen 3.6-Max-Preview, described as its most powerful Qwen model so far, and claims it leads multiple coding/agent benchmarks and improves knowledge and instruction-following versus Qwen 3.6-Plus. The model is offered as a proprietary hosted API (no open weights) via Qwen Studio and Alibaba Cloud Model Studio.","topics":["AI model release","benchmarks","agentic coding","proprietary hosted model","API compatibility","Qwen model lineup","Chinese AI monetization shift"]},"source":"Decrypt RSS","source_domain":"decrypt.co","summary":"Alibaba released Qwen3.6-Max-Preview, its most powerful AI model yet, topping six major coding benchmarks and posting gains in world knowledge and instruction following over its predecessor.","tickers":[],"title":"Alibaba Drops Qwen 3.6 Max Preview\u2014Its Most Powerful Model Yet","url":"https://decrypt.co/364948/alibaba-qwen-3-6-max-preview-most-powerful-model"}}... |