Gate News message, April 29 — Ant Group’s Ling-2.6-flash model weights are now open-sourced, having previously been available only via API. The model features 104 billion total parameters with 7.4 billion activated per inference, a 256K context window, and MIT licensing. BF16, FP8, and INT4 precision versions are available on HuggingFace and ModelScope.
Ling-2.6-flash introduces hybrid linear attention improvements over Ling 2.0, upgrading the original GQA to a 1:7 MLA plus Lightning Linear hybrid architecture combined with highly sparse MoE. Inference efficiency significantly exceeds comparable models: peak generation speed reaches 340 tokens/s on 4x H20 GPUs, with prefill and decode throughput approximately 4x higher than comparable open-source models. Agent-related benchmarks show strong performance: BFCL-V4, TAU2-bench, SWE-bench Verified (61.2%), Claw-Eval, and PinchBench achieve or approach SOTA levels. Across the full Artificial Analysis benchmark suite, total token consumption is only 15 million. On AIME 2026, the model scored 73.85%.
Ant Group’s official website also lists Ling-2.6-1T (trillion-parameter flagship version) and Ling-2.6-mini (lightweight version), though as of publication, their weights remain unreleased on HuggingFace, with only the flash series available for download.
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
Related Articles
OpenAI Models to Gradually Migrate to Amazon's Custom Trainium Chip, Altman Says He's 'Looking Forward' to It
Gate News message, April 29 — OpenAI models running on Amazon Web Services' Bedrock will gradually migrate to Trainium, Amazon's custom-designed AI chip, according to recent remarks from OpenAI CEO Sam Altman and AWS executives. Currently, models operate in a mixed environment using both GPUs and Tr
GateNews15m ago
Altman: Token-Based Pricing Becoming Obsolete as GPT-5.5 Shifts Focus to Task Completion Over Token Count
Gate News message, April 29 — OpenAI CEO Sam Altman said in an interview with Ben Thompson on Stratechery that token-based pricing is not a long-term viable model for AI services. Using GPT-5.5 as an example, Altman noted that while the per-token price is significantly higher than GPT-5.4, the model
GateNews19m ago
Sam Altman posted screenshots of the Codex dual-mode, with office and programming functions officially split.
OpenAI CEO Sam Altman shared screenshots and a statement on X on April 29, and Codex is rolling out a new guided interface. When users enter for the first time, they must choose between two modes: Excelmogging and Codemaxxing. Codex’s current weekly active users have already exceeded 4 million, and its use cases have expanded from code generation to non-technical applications.
MarketWhisper54m ago
OpenAI's Codex Rolls Out Dual-Mode Interface: Excelmogging for Office Work, Codemaxxing for Coding
Gate News message, April 29 — OpenAI CEO Sam Altman announced a redesigned Codex interface on X today, introducing two distinct modes for users. "Excelmogging" targets everyday office tasks with a simplified interface and the tagline "Same tools, simpler interface," featuring example tasks like
GateNews1h ago
US media: A White House draft executive order would allow Anthropic Mythos models to enter government
According to a report citing an insider, quoted by Axios on April 28, the White House is drafting guidance that would allow federal agencies to bypass Anthropic’s Supply Chain Risk Determination (SCRD) and introduce new models for government use, including Anthropic’s Mythos model. In connection with this, the White House issued an official statement saying that any policy statements would be released directly by the President, and that any other claims are purely speculation.
MarketWhisper1h ago
Taylor Swift applies for audio and image trademarks to prevent the spread of AI impersonation content
According to a report by the BBC on April 28, American pop singer Taylor Swift has filed three trademark applications in the United States, covering sound audio clips and stage images. The purpose is to protect her voice and appearance from AI impersonation; trademark attorney Josh Gerben first disclosed the details of the above applications on his blog.
MarketWhisper2h ago
