whatmodel.sh / Ling 3.0 Flash (124B-A5B)

Ling 3.0 Flash (124B-A5B)

+ contribute

Ling 3.0 Flash (124B-A5B) is an open-weights language model from inclusionAI (Ant Group), released August 2026. Its published size is 124B parameters (5.1B active per token) and a 131k-token context window. Licence: mit.

71.2±1.5#73 of 719 LLMsfrom 13 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of Ling 3.0 Flash (124B-A5B), out of 828 with a score, then more from inclusionAI (Ant Group).

#	model	score	where	RAM	released
76	North Small Translate 1.0 Cohere	71±6	api price unknown		2026-08
77	GPT-5.4 mini OpenAI	71±1	api $1.69/M		2026-03
78	Ling 3.0 Flash (124B-A5B) inclusionAI (Ant Group)	71±2	api $0.090/M +local	~75GB	2026-08
79	Mistral Large 4 (public preview) Mistral	71±2	api $2.06/M		2026-10
80	GPT-5.1 OpenAI	71±1	api $3.44/M		2025-11
	more from inclusionAI (Ant Group)
100	Ring-2.6-1T inclusionAI (Ant Group)	69±2	comparison only		2026-05

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks······█····· #1 of 1 · 26% of its score · 3 tests

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Pro (split unspecified) lab	56.6% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 34/55	very hard · avg 9.2%	▇▇▇▇▇▇ 14%	sharp · at its level · small run
SWE-bench Multilingual lab	72.4% ✎	▁··▁▃▂▅▇█▃▃▁ beats 38/60	hard · avg 24%	▇▇▇ 7.2%	at its level · small run
BFCL V4 (December 2025 archive) lab	73.0% ✎	▁▁▁▂▇▆▆▆█▂▄▃ beats 109/111	medium · avg 38%	▇▇ 4.5%	at its level · small run

coding······█····· #1 of 1 · 27% of its score · 3 tests

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Pro (split unspecified) lab	56.6% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 34/55	very hard · avg 9.2%	▇▇▇▇▇▇ 14%	sharp · at its level · small run
SWE-bench Multilingual lab	72.4% ✎	▁··▁▃▂▅▇█▃▃▁ beats 38/60	hard · avg 24%	▇▇▇ 7.2%	at its level · small run
LiveCodeBench 2024-08–2025-05 (version unstated) lab	82.8% ✎	▃···█··▅··▃▃ beats 6/7	medium · avg 48%	▇▇ 5.8%	small run

math······█····· #1 of 1 · 31% of its score · 3 tests

test	result	vs the others in its run	difficulty	counts	why
AIME 2026 lab	93.2% ✎	▁▁▁▂▁▂▆█▆▂·▁ beats 34/49	medium · avg 58%	▇▇▇▇ 11%	sharp · maxed out · small run
IMO-AnswerBench lab	83.7% ✎	▁▁·▁▁▁▄▅█▄▂▃ beats 30/42	medium · avg 38%	▇▇▇▇ 10%	small run
HMMT February 2026 lab	87.0% ✎	▃▃▂▃·██▆█▅▃▅ beats 27/33	medium · avg 40%	▇▇▇▇ 10%	sharp · small run

other tests8 not in a task · 36% of its score

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam lab	22.7% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 53/104	very hard · avg 4.1%	▇▇▇▇ 10%	small run
SkillsBench 87-task subset lab	44.8% ✎	▄··▄▄▄···█·█ beats 4/7	very hard · avg 8.2%	▇▇▇ 7.7%	at its level
WideSearch corrected dataset lab	73.6% ✎	▃······▃▃▃█▃ beats 4/7	hard · avg 20%	▇▇▇ 7.0%	at its level · noisy
SysBench (aggregation unspecified) lab	93.6% ✎	█··█····█··█ beats 3/3	medium · avg 61%	▇▇ 4.6%	small run
Multi-IF (aggregation unspecified) lab	87.7% ✎	▁▁▁▁▁▁▃▄█▄▁▃ beats 40/41	easy · avg 65%	▇▇ 4.0%	blunt · small run
IFBench (aggregation unspecified) lab	74.5% ✎	▃▇▂▄▅▅▂▄▄▆█▇ beats 61/80	hard · avg 34%	▇ 3.2%	at its level · noisy · small run
MCP-Atlas v1 public500 lab	65.5% ✎	█··········█ beats 1/1		—	not counted yet: no other model in the same run to compare it with
BrowseComp single-agent with context management lab	72.2% ✎	█··········█ beats 0/1		—	not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · DeepInfra · $0.090/M · $0.06 in / $0.18 out per 1M tokens · 131k context · deepinfra.com ↗ · price sourceUSD per 1M text tokens. DeepInfra standard synchronous API; model inclusionAI/Ling-3.0-flash. Uncached input; output includes generated reasoning where billed. Provider input limit 131,072 tokens; quantization not specified in the provider record. Batch, cache storage, tools and other modalities excluded.

llama.cpp · recommended

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run Ling 3.0 Flash (124B-A5B) · needs ~74.9 GB · from huggingface.co/inclusionAI (official)
chat in the terminal
llama cli -hf inclusionAI/Ling-3.0-flash-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf inclusionAI/Ling-3.0-flash-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Ling 3.0 Flash (124B-A5B) · needs ~74.9 GB · from huggingface.co/inclusionAI (official)
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/inclusionAI/Ling-3.0-flash-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Ling 3.0 Flash (124B-A5B) · needs ~70 GB, Mac only · from huggingface.co/TensorFold
chat in the terminal
mlx_lm.chat --model TensorFold/Ling-3.0-flash-MLX-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model TensorFold/Ling-3.0-flash-MLX-4bit

Official weights and documentation: huggingface.co/inclusionAI ↗

$ how the score was worked out · inclusionAI/Ling-3.0-flash mo… 99%

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
inclusionAI/Ling-3.0-flash model card · lab-reported	13	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%	MIT
prior · pull toward the middle		▇ 1%
weighted average		= 71.2

not counted (2)

inclusionAI/Ling-3.0-flash model card	MCP-Atlas v1 public500 [correct]	65.5%	fewer than three models have this test yet, or this source ran it for this model alone
inclusionAI/Ling-3.0-flash model card	BrowseComp single-agent with context management [correct]	72.2%	fewer than three models have this test yet, or this source ran it for this model alone

$ 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