Tasks and inputs
Leanstral 2603 is listed for Coding & Code Generation. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Oct 7, 2026
Leanstral 2603 is an Apache-licensed model for Lean 4 formal proofs and code 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
Leanstral 2603 is an Apache-licensed model for Lean 4 formal proofs and code workflows.
Leanstral is intended for proof and coding tasks where an external checker can establish whether an artifact is valid. The publisher describes 119 billion total parameters, about 6.5 billion active parameters and a 256K hard context window, with a recommendation to stay at or below 200K. This is the earlier Leanstral checkpoint, useful as a reproducible baseline when assessing changes in the later release.
Give the model a theorem, the permitted imports and the exact Lean project version. Check that the generated proof compiles and that it proves the requested statement without adding an unintended assumption. Include a task requiring a small proof repair and another that needs selecting an appropriate lemma. A fluent explanation is useful for a reader, but acceptance should depend on the actual proof artifact.
Use the model’s documented chat template and reasoning settings. The model supports image input, but visual input should be evaluated within a specific workflow rather than inferred from a general coding result. Local serving requires the complete expert weights, not only the active parameter footprint. Managed access through publisher tools depends on the available account and model route. When comparing with a hosted Leanstral listing, preserve the distinction between the downloadable checkpoint and the host’s endpoint, limits and prices.
Compare the model on the work you need to complete.
Leanstral 2603 is listed for Coding & Code Generation. It accepts text, image 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 Apache 2.0. 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.
inclusionAI
Ling 2.6 1T is a trillion-parameter MIT-licensed checkpoint for complex instruction following, coding and long-document 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.