whatmodel.sh / LongCat-Flash-Thinking (560B-A27B)

LongCat-Flash-Thinking (560B-A27B)

+ contribute

LongCat-Flash-Thinking (560B-A27B) is an open-weights language model from Meituan, released September 2025. Its published size is 560B parameters (27B active per token) and a 128k-token context window. Licence: mit.

Catalogue identity notes

Meituan LongCat team, meituan-longcat/LongCat-Flash-Thinking @7c9aad1 (created 2025-09-21, MIT). Comparison-only for now: weights are public (about 1.12 TB BF16, 75 safetensors files) but the card points to an external SGLang/vLLM deployment guide rather than giving a command, so no run option is recorded. Its own columns sit in the card runs that already hold six catalogued comparators. Reasoning model (arXiv 2509.18883). Not LongCat 2.0 or the 2602-exp chat preview (separate entries).

Model information sources

Release date ↗ · Weights licence ↗

Card: '560 billion total parameters ... activates 18.6B∼31.3B parameters (averaging∼27B) based on contextual demands'; safetensors metadata 561,862,880,256.

68.5±1.0comparison only: no current run option verifiedfrom 18 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of LongCat-Flash-Thinking (560B-A27B), out of 828 with a score, then more from Meituan.

#	model	score	where	RAM	released
112	GLM-5V-Turbo Z.ai	69±2	api $1.90/M		2026-04
113	GPT-5.2-Codex OpenAI	69±2	comparison only		2026-01
114	LongCat-Flash-Thinking (560B-A27B) Meituan	69±1	comparison only		2025-09
115	Step 3.5 Flash (196B-A11B) StepFun	68±1	api $0.15/M +local	~118GB	2026-02
116	DeepSeek V3.2 DeepSeek	68±1	api $0.29/M +local	~110GB	2025-12
	more from Meituan
63	LongCat 2.0 (1.6T-A48B) Meituan	73±2	api $1.30/M +local	~2155GB	2026-06
157	LongCat-Flash-Chat (560B-A27B) Meituan	66±1	local	~333GB	2025-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▂▆█▆▅▄▄▃▂▁ #104 of 340 · 16% of its score · 4 tests

Beats Grok 4.20 (0309 reasoning), Grok 4.3, o3-pro here, though they rank higher overall. Trails SmolLM3 3B, Nova 2 Pro (Preview), MiniMax-M2.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
τ²-bench Retail lab	71.5% ✎	█···█·█·██·█ beats 4/5	medium · avg 36%	▇▇▇ 7.2%	at its level
τ²-bench Telecom lab	83.1% ✎	█··██·█·█··█ beats 5/5	very hard · avg 10%	▇▇ 4.7%	sharp · at its level · noisy
τ²-bench Airline lab	67.5% ✎	▄·····█··▄·█ beats 5/5	hard · avg 29%	▇ 3.6%	at its level
BFCL v3 lab	74.4% ✎	▁▁▁▁▁▃▅█▆▅▄▁ beats 165/170	medium · avg 56%	▇ 0.9%	blunt · at its level

coding▁▁▁▃▅▇█▇▅▄▃▁ #57 of 524 * · 5.9% of its score · 1 test

Beats GPT-5.4, DeepSeek V4 Flash Vision (Exp), Gemini 3.1 Pro (Preview) here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench (version unspecified) lab	79.4% ✎	▄······▄█▄·▄ beats 5/5	medium · avg 39%	▇▇ 5.9%	typical

chat & writing▁▁▁▃▅▄▆█▇▇▇▃ #147 of 463 * · 3.0% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B) here, though they rank higher overall. Trails LFM2.5 8B-A1B, LFM2.5 2.6B, Nemotron-Cascade-8B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
IFEval prompt-level strict lab	86.9% ✎	███··█·█···█ beats 2/5	easy · avg 74%	▇ 3.0%	typical

math▁▂▃▅▇▅▅█▄▁▁▁ #76 of 462 · 30% of its score · 4 tests

Beats Claude Sonnet 4.6, Claude Opus 4.5, Qwen3.7 Plus here, though they rank higher overall. Trails Granite 4.2 30B, INTELLECT-3, EXAONE 4.5 33B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2025 lab	90.6% ✎	▁·▁▁▁▂▅▆█▇▅▁ beats 117/133	medium · avg 51%	▇▇▇▇ 10%	sharp
AIME 2024 lab	93.3% ✎	▁▁▁▂▁▃▁▅███▃ beats 82/85	medium · avg 60%	▇▇▇▇ 9.0%	sharp · maxed out
HMMT 2025 (edition unspecified) lab	83.7% ✎	▄··▄··▄·▄··█ beats 4/5	hard · avg 29%	▇▇▇ 8.4%	sharp
MATH-500 lab	99.2% ✎	█···██·█·█·█ beats 4/5	very easy · avg 91%	▇ 2.4%	maxed out · noisy

other tests8 not in a task · 44% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond lab	81.5% ✎	▄··▄··▄▄···█ beats 2/5	hard · avg 43%	▇▇▇▇▇ 14%	sharp
MMLU-Pro lab	82.6% ✎	▄·▄···█·▄··▄ beats 1/5	medium · avg 56%	▇▇▇ 7.0%	typical
MMLU-Redux lab	89.3% ✎	████··█····█ beats 0/5	medium · avg 73%	▇▇ 5.9%	typical
VitaBench lab	29.5% ✎	█····██·██·█ beats 4/5	very hard · avg 7.8%	▇▇ 5.0%	at its level
OJBench lab	40.7% ✎	▄······▄▄·▄█ beats 4/5	very hard · avg 12%	▇▇ 4.7%	at its level
ZebraLogic lab	95.5% ✎	██··█·██···█ beats 3/5	easy · avg 75%	▇▇ 4.5%	maxed out
BeyondAIME lab	69.5% ✎	█··▄····▄··█ beats 3/5	medium · avg 41%	▇ 2.9%	at its level
miniF2F (pass@1) lab	67.6% ✎	██······▄··▄ beats 5/5	very hard · avg 13%	▇ 0.4%	blunt · at its level · noisy

$ where to run it

No verified API or download recorded yet · add one

Weights and documentation: huggingface.co/meituan-longcat ↗

$ how the score was worked out · Meituan LongCat-Flash-Thinkin… 100%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Meituan LongCat-Flash-Thinking model card (evaluation table) · lab-reported	18	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 100%	MIT
prior · pull toward the middle		▇ 0%
weighted average		= 68.5

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json