Benchmark Almanac

Spark-X2.6-4B

Spark-X2.6-4B.

iFlytek Released October 22nd, 2026 Family Spark X2.6 The lab's model card Compare with others
26
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
iFlytek Oct 2026

Spark-X2.6-4B

Open weights Small

142.4 #159 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • IFBench 75.8% #23 /80
  • SciCode 36.8% #37 /49
  • BFCL V4 71.8% #6 /30

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
IFEval 93.6% #13 of 135 Lab model card
IFBench 75.8% #23 of 80 Lab model card
Humanity's Last Exam 36.9% #69 of 154 Lab model card
LongBench v2 53.3% #22 of 35 Lab model card
GraphWalks 63.2% #20 of 27 Lab model card
OpenAI MRCR v2 45.8% #12 of 14 Lab model card
AA-LCR 1.1 67.3% #6 of 6 Lab model card

Math and science

BenchmarkScoreRankSource
GPQA Diamond 77.1% #137 of 302 Lab model card
AIME 2026 91.2% #26 of 38 Lab model card
IMO-AnswerBench 75.2% #33 of 37 Lab model card
HMMT 2026 81.3% #17 of 19 Lab model card
Gaokao Math 2026 134.9 #1 of 1 Lab model card

Coding

BenchmarkScoreRankSource
SWE-Bench Pro 54.3% #54 of 81 Lab model card
SWE-bench Verified 62.8% #104 of 155 Lab model card
SWE-bench Multilingual 64.7% #54 of 73 Lab model card
SciCode 36.8% #37 of 49 Lab model card

Agents and tools

BenchmarkScoreRankSource
BFCL V4 71.8% #6 of 30 Lab model card
tau2-bench 86.9% #23 of 79 Lab model card
Claw-Eval 69.9% #6 of 13 Lab model card
EnterpriseOps-Gym 33.6% #2 of 3 Lab model card
BrowseComp 53.1% #61 of 81 Lab model card
MCP Atlas 62.4% #35 of 46 Lab model card
tau3-bench 34.7% #12 of 14 Lab model card
PAgentBench 74.1% #1 of 1 Lab model card
SWebBench 71.7% #1 of 1 Lab model card
VitaBench 2.0 30.8% #1 of 1 Lab model card

Its lab's numbers put it at 142.4, number 159 of 404

The Self-Reported Index fits one score to the 20 capability results iFlytek reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 140.0 to 144.9. How we compute it.

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Spark-X2.6-4B: Advancing the Frontier of On-Device Agents for Real-World Tasks (Hugging Face) Oct 22, 2026