Tasks and inputs
Ling 2.6 1T is listed for Text generation, Coding & Code Generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Oct 8, 2026
Ling 2.6 1T is a trillion-parameter MIT-licensed checkpoint for complex instruction following, coding and long-document workflows.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Ling 2.6 1T is a trillion-parameter MIT-licensed checkpoint for complex instruction following, coding and long-document workflows.
Ling 2.6 1T combines MLA and linear attention in a sparse expert architecture. The publisher positions it for coding and everyday agent workflows, with training intended to reduce redundant reasoning output. Its released configuration supports 262,144 positions.
A trillion-parameter checkpoint is a cluster deployment project even when only selected experts activate for a token. Plan weight storage, accelerator memory, networking, runtime support and the cache needed for concurrent long conversations. Use the publisher’s distributed serving recipe as the starting point for a capacity test.
Give the assistant several related documents or source files and a precise completion criterion. Include a later tool result that changes an earlier assumption. Review whether the final answer reflects the updated evidence and retains constraints from the beginning of the conversation. Track output overhead and rejected attempts, not only tokens per second.
The model card identifies the MIT licence. This profile covers the downloadable checkpoint and keeps hosted token prices unset. A routed API entry for the same family may use a different identifier, context limit or price. Record the exact route or revision used in an evaluation so checkpoint performance and service behaviour are not mixed together.
Compare the model on the work you need to complete.
Ling 2.6 1T is listed for Text generation, Coding & Code Generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 262.14K context. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under MIT. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Mistral AI
Leanstral 1.5 119B A6B is an Apache-licensed model for Lean 4 formal proofs and code workflows.
Mistral AI
Leanstral 2603 is an Apache-licensed model for Lean 4 formal proofs and code workflows.
inclusionAI
Ling 2.6 Flash is a hybrid-attention instruction checkpoint designed for concise responses and repeated agent tool workflows.
inclusionAI
Ling 3.0 Tiny is a compact MIT-licensed reasoning checkpoint with 7.9B total parameters and 1.3B active per token.
Meituan LongCat
LongCat 2.5 Preview is Meituan’s hosted multimodal reasoning model for coding and agent work, with a one-million-token context.
Meituan LongCat
LongCat Flash Lite Sparse adds sparse attention and native long-context support to the non-thinking LongCat Lite family.