Skip to main content

← Dracon Bench leaderboard

Model guide · Dracon Bench

ling-3.2-flash

Tested First seen 2026-11-13 (listed on the provider) · First tested 2026-11-17 · OpenRouter free lane

InclusionAI Ling 3.2 Flash ranks eleventh with five scored suites; the NIM rows added in November 2026 moved it down from seventh without changing its scores. It matches the pilot best on IFEval-50 at 90.0% (n=10) and passes all five JSON briefs, but the five SVG briefs all timed out, so no SVG score is shown.

Rank
#11
Overall
64.0
no public benchmark yet
Checked the OpenRouter catalogue, Artificial Analysis and the public score trackers on 2026-11-22; none lists a score for this model.
Suites
5 of 6
Ranked
#11 of 21

When the world heard about it

Timeline

  1. First seen on the provider 2026-11-13 the provider's own catalogue
  2. First tested by us 2026-11-17 Dracon Bench
  3. Officially announced 2026-11-13 models.dev
  4. First official benchmark not recorded no source lists it
  • Reachable before we scored it 4 days
  • Reachable before the vendor announced it same day

Track A · script-scored subsets

Scores

Subset scores for ling-3.2-flash.
SuiteScoreSamplePrecision
IFEval-50 prompt-strict90.0%9/10 prompts59.6%–98.2%
IFEval-50 instruction-strict94.1%16/17 constraints73.0%–99.0%
IFBench-50 prompt-strict40.0%4/10 prompts16.8%–68.7%
IFBench-50 instruction-strict36.4%4/11 constraints15.2%–64.6%
MMLU-Pro-5020.0%2/10 served5.7%–51.0%
JSON briefs conformance100.0%5/5 served56.6%–100.0%
GSM8K-5070.0%7/10 served39.7%–89.2%

The range beside a script-scored figure is a 95% interval over that figure's own sample: the prompts it followed, the constraints it met, or the items it answered. The instruction-strict figures are measured over CONSTRAINTS, which is why their sample line counts constraints rather than prompts. The judged total is not a proportion, so it has no interval: it carries the measured judge band instead: how far a re-judging of the same artifacts moved them, which is a different kind of number. The method page says what each one covers and what it does not.

Track A · per-type breakdown

What it failed

The instruction types these two suites check, and the checks the model missed. A type is checked once per prompt that carries it, so a type with seven checks was asked for seven times. The counts beside a type are that type's own; the suite's totals are in the scores table above. These are the failures behind the instruction-strict figure, which is the sum of them.

IFEval-50

Missed at least one check in 1 of the 15 instruction types checked.

The instruction types 1 of 15 the IFEval-50 suite checked that inclusionai/ling-3.2-flash:free missed, most missed first.
Instruction typeChecksMissed
detectable format · json format11

IFBench-50

Missed at least one check in 7 of the 8 verifiers checked.

The verifiers 7 of 8 the IFBench-50 suite checked that inclusionai/ling-3.2-flash:free missed, most missed first.
VerifierChecksMissed
count · conjunctions11
count · numbers21
count · unique word count31
count · word count range11
format · list11
words · keywords specific position11
words · words position11

Provider-reported usage

Thinking cost

What the run cost the model in thinking: provider-reported completion tokens per prompt, the thinking share, and the cost per correct answer. Reasoning-heavy models can exhaust fixed budgets before answering.

Thinking cost for ling-3.2-flash.
SuiteCostReading
IFEval-50957/prompt46% thinking· 1,063/correct
GSM8K-50364/prompt67% thinking· 521/correct