Beating AI Flash News: InclusionAI has released its new Flash model, Ling-3.1-flash. The model has approximately total parameters, with only about activated per token token, and can scale up to a context. Compared with Ling-3.0-flash's total parameters and 5.1B activated parameters, the total scale of this generation has increased by approximately 4.5 times, while the activated parameters per token token are nearly 5 times larger.
InclusionAI also highlighted its long-duration coding capabilities this time. According to the official statement, Ling-3.1-flash continuously spent approximately 17 hours writing a Lua-to-x86-64 ELF compiler from scratch, with 182 tests passing 178 of them, for a success rate of 97.8%. In another task lasting approximately 20 hours, it ported a C image library to Rust, ultimately achieving 8.015× acceleration, with all 30 correctness checks passed.
The model can now be tried through Novita AI and Vercel AI Gateway. The free trial will continue for 2 weeks. Currently, only approximately context is available; the full has not yet been launched. InclusionAI stated that after transitioning to paid services, it will make the context available and plans to open-source the model simultaneously.
Original link https://m.theblockbeats.info/flash/369865
Key points
01
InclusionAI released the Flash model Ling-3.1-flash。
02
The report states that the model has approximately total parameters, with only about activated per token token, and can scale up to a context.
03
Compared with Ling-3.0-flash's total parameters and 5 active parameters, reports indicate that the new model's total scale has increased by approximately 4.5 times, while its single token active parameters are nearly 5 times greater.
04
According to the official statement, Ling-3.1-flash continuously spent approximately 17 hours writing a Lua-to-x86-64 ELF compiler from scratch, with 182 tests passed 178 tests, for a success rate of 97.8%。
05
According to the official statement, in another task lasting approximately 20 hours, the model ported a C imaging library to Rust, achieving 8.015 times acceleration, with 30 correctness checks all passed.
06
The model can be tried through Novita AI and Vercel AI Gateway; the free trial will last for 2 weeks.
07
The current trial is limited to approximately context, and the full context is not yet available.
08
Bailing says that context will be made available after the service becomes paid, and the model is also planned to be open-sourced at the same time.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This is a newly released AI model. Reports focus on its parameter scale, the context length it can process at one time, and two officially demonstrated long-duration programming tasks. The currently available context is approximately; the claimed full capability is not yet available.
Why it matters to readers
If the reported long-duration coding performance and context capabilities can be verified, this model may attract the attention of developers who need to handle longer coding tasks; the available information can currently only confirm that these claims have been reported, not whether the tests underwent independent review.
Beginners can understand it as a AI model capable of generating and processing code. The number of tests passed, runtime, and speedup multiple are results officially demonstrated by the developers and do not mean that all real-world tasks can achieve the same performance.
Risks and unknowns
No official original announcement from Ant Bailing has been provided.
Compiler tests, Rust porting results, and 8.015-fold speedup have not been independently verified.
The full context is not yet available, and its actual usability cannot be confirmed from the available materials.
Making the context and open-source models available after payment are both future plans; whether and when they will be implemented remains uncertain.
Free trial 2 week's specific start and end times are not stated in the evidentiary text.
Related Developments
Loading event timeline…
Related concepts
Total parameters
This term is not in the glossary yet. Browse related concepts in the glossary.
