Ling-3.1-flash
A large-scale Mixture-of-Experts (MoE) language model developed by the inclusionAI team, with approximately 560 billion total parameters and approximately 25 billion parameters activated per token. The model supports a context window of up to one million tokens.
Timeline (last 90 days)
Ant Group's Ling-3.1-flash model is about to receive an update codenamed "1019."
The announcement did not give a specific release date for the Ling-3.1-flash 1019 version.
The Ant Ling announcement marked both API access and open-source weights for Ling-3.1-flash 1019 as "coming soon."
AutomationBench-AA's 68% completion rate indicates that models still have about one-third of failure space in automated tasks, and there is still a gap to reliable production-grade agents.
The Ling-3.1-flash 1019 version raised its Elo score on AA-Briefcase from about 1400 to 1524.
The Ling-3.1-flash 1019 version improved its completion rate on AutomationBench-AA from 62% to 68%.
The Ling-3.1-flash 1019 version improved its score on the Artificial Analysis Intelligence Index from 41 to 44.
inclusionAI's Ling-3.1-flash takes first place on SWE-bench_Pro (68.74), SWE-bench_Multilingual (88.5), and skillsbench (68.7), while zai-org's GLM series holds first place on evaluations such as ResearchClawBench, EvasionBench, and chi-bench.
Ling-3.1-flash scored 68.70 on the SkillsBench benchmark.
Ling-3.1-flash releases open weights under the MIT license.
13 more timeline events
All Facts (20)
The Ling-3.1-flash 1019 version improved its score on the Artificial Analysis Intelligence Index from 41 to 44.
50%UnverifiedThe announcement did not give a specific release date for the Ling-3.1-flash 1019 version.
50%UnverifiedThe Ant Ling announcement marked both API access and open-source weights for Ling-3.1-flash 1019 as "coming soon."
50%UnverifiedAutomationBench-AA's 68% completion rate indicates that models still have about one-third of failure space in automated tasks, and there is still a gap to reliable production-grade agents.
50%UnverifiedThe Ling-3.1-flash 1019 version raised its Elo score on AA-Briefcase from about 1400 to 1524.
50%UnverifiedThe Ling-3.1-flash 1019 version improved its completion rate on AutomationBench-AA from 62% to 68%.
50%UnverifiedinclusionAI's Ling-3.1-flash takes first place on SWE-bench_Pro (68.74), SWE-bench_Multilingual (88.5), and skillsbench (68.7), while zai-org's GLM series holds first place on evaluations such as ResearchClawBench, EvasionBench, and chi-bench.
50%UnverifiedOfficial documentation and serving examples for tool calling are provided.
50%UnverifiedLing-3.1-flash scored 1673 on the GDPval-AA v2.1 benchmark.
50%UnverifiedLing-3.1-flash scored 52.50 on the Automationbench Public benchmark.
50%UnverifiedLing-3.1-flash scored 68.70 on the SkillsBench benchmark.
50%UnverifiedThe source material does not provide comparison targets for these scores, so it is impossible to judge where the 1673 score stands among comparable models.
50%UnverifiedThe top of terminal-bench-2.0 is mindlab-research/Macaron-V1-Venti (87.6), terminal-bench-2.1 belongs to DeepSeek-V4.1-Flash (90.6), and the base version terminal-bench is led by Ling-3.1-flash with 46.03.
50%UnverifiedLing-3.1-flash adopts a Mixture of Experts (MoE) architecture with a total parameter count of approximately 560 billion (~560B).
50%UnverifiedLing-3.1-flash claims to support a context window of up to 1 million tokens (1M).
50%UnverifiedLing-3.1-flash activates only approximately 25 billion (~25B) parameters per token.
50%UnverifiedAnt Group's Ling-3.1-flash model is about to receive an update codenamed "1019."
50%UnverifiedThe publisher demonstrated a hands-on case: uploading a CSV file, which the model turns into an interactive financial analysis page.
50%UnverifiedLing-3.1-flash releases open weights under the MIT license.
50%UnverifiedAnt Group launched Ling-3.1-flash, with approximately 560 billion total parameters, using a MoE architecture that activates only approximately 25 billion parameters per token, with a context limit of 1 million.
50%