Data as of October 25, 2026 · How the score is built
Released Jul 31, 20261M contextMeituan LongCat-Flash-Lite-Sparse model card
LongCat-Flash-Lite-Sparse
Decision readingLongCat-Flash-Lite-Sparse is tracked, but not publicly ranked yet. The profile exposes 17 sourced benchmark rows and leaves unsupported fields blank until a published record exists.
Released Jul 31, 2026 — see all recent releases
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Capability
Unranked
field median 50Not eligible for a public rank
Price
Self-hosted; infrastructure cost varies
input median $0.75No comparable first-party hosted token rate
Speed
Not measured
field median 96 tok/sTime to first token not measured
Context
1Mtokens
field median 256,000Maximum output length is tracked separately
Strongest published evidence
Published rows are visible, but no category has enough eligible evidence for a comparative rank.
Validate before choosing
17 published rows leave some tracked benchmark slots empty. No comparable first-party API token rate is published.
Source-linked · 17 displayable benchmark rows
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticRank Not rankedWeight 22%4 benchmarksVerified | 14.1 | 4 benchmarks | Verified | |||
| CodingRank Not rankedWeight 20%4 benchmarksVerified | 29.8 | 4 benchmarks | Verified | |||
| ReasoningWeight 17%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MultimodalWeight 12%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| KnowledgeRank Not rankedWeight 12%5 benchmarksVerified | 74.2 | 5 benchmarks | Verified | |||
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| Inst. FollowingWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MathRank Not rankedWeight 5%4 benchmarksVerified | 33.1 | 4 benchmarks | Verified |
17 of 766 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic4/4 verified
- Coding4/4 verified
- ReasoningNot measured
- MultimodalNot measured
- Knowledge5/5 verified
- MultilingualNot measured
- Inst. FollowingNot measured
- Math4/4 verified
Benchmark ledger
Coding opens by default. The marker compares each value with the best source-verified result in the catalog; provisional leaders do not set the reference. Expand the remaining categories for every published row.
Coding4 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| SWE-bench Pro | Score40.6% | Versus best verified row Best verified: Claude Opus 5.5 · 89.9% | Gap49.3 behind | WeightDisplay only | Provider exact |
| SWE Multilingual | Score59.3% | Versus best verified row Best verified: Claude Opus 5.5 · 93.9% | Gap34.6 behind | WeightDisplay only | Provider exact |
| SWE-bench VerifiedSoftware Engineering Benchmark Verified | Score68.2% | Versus best verified row Best verified: Claude Opus 5 · 96% | Gap27.8 behind | WeightDisplay only | Provider exact |
| Terminal-Bench 2.0 | Score33.7% | Versus best verified row Best verified: GPT-5.5 · 82.0% | Gap48.3 behind | WeightDisplay only | Provider exact |
Agentic4 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| BrowseComp | Score48.6% | Versus best verified row Best verified: Atria Dawn Preview · 92.5% | Gap43.9 behind | WeightDisplay only | Provider exact |
| Terminal-Bench 2.0 | Score33.7% | Versus best verified row Best verified: GPT-5.5 · 82% | Gap48.3 behind | WeightDisplay only | Provider exact |
| MCP Atlas | Score45.6% | Versus best verified row Best verified: Muse Spark Open · 89.2% | Gap43.6 behind | WeightDisplay only | Provider exact |
| VITA-Bench | Score21.7% | Versus best verified row Best verified: Qwen3.7 Max · 47.9% | Gap26.2 behind | WeightDisplay only | Provider exact |
Knowledge5 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| MMLU-ProMassive Multitask Language Understanding Professional | Score79.2% | Versus best verified row Best verified: Qwen3.7 Max · 89.6% | Gap10.4 behind | WeightDisplay only | Provider exact |
| MMLUMassive Multitask Language Understanding | Score85.3% | Versus best verified row Best verified: o1 · 91.8% | Gap6.5 behind | WeightDisplay only | Provider exact |
| CMMLUChinese Massive Multitask Language Understanding | Score84.3% | Versus best verified row Best verified: LongCat-Flash-Lite-Sparse · 84.3% | GapBest verified | WeightDisplay only | Provider exact |
| C-Eval | Score85.8% | Versus best verified row Best verified: Qwen3.6 Plus · 93.3% | Gap7.5 behind | WeightDisplay only | Provider exact |
| GPQA-DGPQA Diamond | Score69.5% | Versus best verified row Best verified: GPT-6 Astra · 96.0% | Gap26.5 behind | WeightDisplay only | Provider exact |
Math4 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| AIME26AIME 2026 | Score65.7% | Versus best verified row Best verified: GLM-5.2 · 99.2% | Gap33.5 behind | WeightDisplay only | Provider exact |
| HMMT Feb 2026Harvard-MIT Mathematics Tournament February 2026 | Score40.5% | Versus best verified row Best verified: Qwen3.7 Max · 97.1% | Gap56.6 behind | WeightDisplay only | Provider exact |
| MATH-500MATH-500 Problem Set | Score95.8% | Versus best verified row Best verified: Ternary Bonsai 2 27B · 98.8% | Gap3 behind | WeightDisplay only | Provider exact |
| IMOAnswerBench | Score49.4% | Versus best verified row Best verified: dots3-note Preview · 90.9% | Gap41.5 behind | WeightDisplay only | Provider exact |
Bars run 0–100; the dark tick marks the best source-verified value
All 17 rowsLineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Sparse
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- meituan-longcat/LongCat-Flash-Lite-SparseMeituan LongCat-Flash-Lite-Sparse model card
- Context window
- 1MMeituan LongCat-Flash-Lite-Sparse model card
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- Not sourced yet
- Output modalities
- Not sourced yet
- Parameters
- Not sourced yet
- Availability
- Meituan publishes the checkpoint under MIT on Hugging Face. It replaces dense MLA in LongCat-Flash-Lite with LongCat Sparse Attention and natively supports a 1M-token context.
- Cloud regions
- Not tracked yet
- Lifecycle
- Current
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Not documented in the pricing recordMeituan LongCat-Flash-Lite-Sparse model card
- Self-host
- LongCat-Flash-Lite-Sparse needs ~48GB VRAM (2× NVIDIA RTX 4090 (24GB)).
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
We track LongCat-Flash-Lite-Sparse, but the public leaderboard excludes this profile until enough non-generated benchmark coverage is available. Only published rows appear above.
LongCat-Flash-Lite-Sparse is a open weight model with a 1M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Meituan publishes the checkpoint under MIT on Hugging Face. It replaces dense MLA in LongCat-Flash-Lite with LongCat Sparse Attention and natively supports a 1M-token context.
Official exact-value snapshot from the LongCat-Flash-Lite-Sparse Hugging Face model card (July 31, 2026), using the without-HI column that Meituan presents first. BenchLM maps SWE-Bench Verified, SWE-Bench Pro, SWE-Bench Multilingual, TerminalBench 2.0 (provider-run, blocklisted), τ²-Telecom is not mapped because the τ² lane is the full average, VitaBench, MCP-Atlas, BrowseComp, MMLU, MMLU-Pro, CMMLU, C-Eval, GPQA-Diamond, MATH500, AIME 2026, HMMT 2026 Feb, and IMO AnswerBench onto existing lanes. RWSearch, BrowseComp-zh, BeyondAIME, and the long-context suite stay outside the schema.
17 of 766 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Last updated October 25, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Questions
How does LongCat-Flash-Lite-Sparse perform overall in AI benchmarks?
LongCat-Flash-Lite-Sparse has 17 source-displayable benchmark rows, but it does not qualify for a public overall rank. The available rows remain visible by category without being converted into a site-wide score. Missing evidence stays blank instead of being estimated from an earlier model.
Is LongCat-Flash-Lite-Sparse good for knowledge and understanding?
LongCat-Flash-Lite-Sparse has source-displayable benchmark coverage for knowledge and understanding, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is LongCat-Flash-Lite-Sparse good for coding and programming?
LongCat-Flash-Lite-Sparse has source-displayable benchmark coverage for coding and programming, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is LongCat-Flash-Lite-Sparse good for mathematics?
LongCat-Flash-Lite-Sparse has source-displayable benchmark coverage for mathematics, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is LongCat-Flash-Lite-Sparse good for agentic tool use and computer tasks?
LongCat-Flash-Lite-Sparse has source-displayable benchmark coverage for agentic tool use and computer tasks, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is LongCat-Flash-Lite-Sparse open source?
LongCat-Flash-Lite-Sparse is an open-weight model from Meituan. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Does LongCat-Flash-Lite-Sparse have full benchmark coverage on BenchLM?
No. LongCat-Flash-Lite-Sparse currently has 17 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of LongCat-Flash-Lite-Sparse?
LongCat-Flash-Lite-Sparse has a documented context window of 1M. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.