Inference
EN
SparkLLM

Spark-X2.6-4B

Spark-X2.6-4B is a reasoning model from SparkLLM in the Spark X2.6 family. 19 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-10
RankedEnough results for an index.
IndexAll categories, out of 100.48.7±5.2Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.—Median 67/s
First tokenSeconds to the first output token.—Median 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.1MMedian 262K
EloLMArena rating and rank.—Median 1422

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.7 counted48.8138 / 218
CodingCode writing and repair.3 counted48.8146 / 209
ReasoningLogic problems and puzzles.2 counted48.5127 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.2 counted43.5177 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted55.822 / 63
MathMath problems.3 counted50.585 / 168

Results

19 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Instruction-Following EvalInstruction93.6%13 / 70
Claw-EvalAgenticAbove fitFar above this model's index.69.9%10 / 42
Berkeley Function Calling Leaderboard v4Agentic71.8%7 / 26
τ²-Bench Tool-Agent-User EvaluationAgentic86.9%49 / 149
Instruction Following BenchmarkInstruction75.8%28 / 57
Artificial Analysis Long Context ReasoningReasoning67.3%129 / 217
Graduate-Level Google-Proof Q&AKnowledge69.9%93 / 141
SWE-bench ProCoding54.3%57 / 83
MCP AtlasAgentic62.4%36 / 52
τ³-Bench Tool-Agent-User EvaluationAgentic34.7%23 / 32
Software Engineering Benchmark VerifiedCoding62.8%82 / 114
AIME 2026Math91.2%25 / 34
Scientific Code BenchmarkCoding36.8%26 / 35
LongBench v2Reasoning53.3%19 / 25
Harvard-MIT Mathematics Tournament February 2026Math81.3%21 / 27
GPQA Diamond · BenchLMKnowledge77.1%66 / 84
IMOAnswerBenchMath75.2%8 / 10
BrowseCompAgentic53.1%56 / 63
Humanity's Last Exam with toolsAgentic36.9%33 / 36
Humanity's Last Exam without toolsKnowledge17.9%50 / 53

Sources

BenchLMCC BY-NC 4.0

History

5 releases
Index49.9 → 48.7
23 Oct 202612 Nov 2026

Questions

How good is Spark-X2.6-4B?
Its index is 48.7 out of 100. That is place 157 of 283 ranked models.
What is Spark-X2.6-4B best at?
Instruction: 55.8 out of 100, place 22 of 63.
What is the context window of Spark-X2.6-4B?
1M tokens.
Is Spark-X2.6-4B open source?
Yes. Its weights are published.
When was Spark-X2.6-4B released?
October 2026.

Badge

Spark-X2.6-4B on Inference360: place 157 of 283, index 48.7
HTML
<a href="https://inference360.ai/models/spark-x2-6-4b"><img src="https://inference360.ai/badges/spark-x2-6-4b.svg" alt="Spark-X2.6-4B on Inference360: place 157 of 283, index 48.7" height="20"></a>
Markdown
[![Spark-X2.6-4B on Inference360: place 157 of 283, index 48.7](https://inference360.ai/badges/spark-x2-6-4b.svg)](https://inference360.ai/models/spark-x2-6-4b)

Compare with

K-EXAONE 2.048.8