Vascend
HorizonVascend IDEVascend CLICommunityBenchmarkContact
Sign inJoin Discord
HomeHorizonIDECLISign in
ChangelogIntelligenceCost per taskCoding Agent IndexCyber IndexDecision IndexBenchmarksSpeedLatencyPriceLeaderboard
Benchmark

Step 5 Preview

  • StepFun
  • Release · September 18, 2026
  • Open weights
  • Reasoning

Intelligence

Intelligence Index

#33 of 266
  1. 45.846GPT-6 Astra (Low)
  2. 45.445Qwen3.8 Max (0902)
  3. 45.145DeepSeek V4.1 Flash 1017 (Max)
  4. 44.845GLM-5.3 (Max)
  5. 43.944Ling 3.1 Flash 1019
  6. 43.744Step 5 Preview
  7. 43.443Claude Haiku 5.5 (Max)
  8. 42.743MiniMax-M3.1-Flash
  9. 42.342Claude Opus 5.5 (Low)
  10. 42.242Grok 4.7 (Low)
Data from Artificial Analysis

Evaluations

  • AA-Briefcase1,423#18
  • AA-Omniscience Accuracy41.5%#52
  • AA-Omniscience Non-Hallucination57%#15
  • AutomationBench-AA51%#20
  • GDPval-AA1,587#16
  • Humanity's Last Exam46.5%#31
  • GDP.pdf14.8%#22
  • Terminal-Bench 4.033.3%#32
  • AA-LCR88.3%#1
  • SciCode58.9%#14
  • CritPt20.9%#35
  • Terminal-Bench-Science2.4%#21
  • FrontierCode38.4%#19
  • AA-AnalystAgent35%#17
  • EnterpriseOps-Gym-AA47.2%#2
  • AA-3D854#16
  • ITBench-AA55.6%#1
  • MLCR-AA18.3%#12
  • MMMU-Pro76.4%#42
  • PersonalityBench: Expressiveness30.7#21
  • PersonalityBench: Organization95#2
  • PersonalityBench: Transparency35.6#20
  • PersonalityBench: Slop45.6#13
  • PersonalityBench: Informativeness45.6#17
  • PersonalityBench: Warmth35.6#20
  • PersonalityBench: Conviction55.5#14
  • PersonalityBench: Imagination38.1#20
  • APEX-Agents38%#3
Step 5 PreviewMedian
Data from Artificial Analysis

Cost per task

Cost per task

$ per Intelligence Index task · #12 of 24
  1. $0.260.26DeepSeek V4.1 Flash 1017 (Max)
  2. $0.280.28GLM-5.3-Flash
  3. $0.720.72GPT-6.1 Sol (Max)
  4. $0.840.84Muse Spark Open (Xhigh)
  5. $1.011.01Qwen3.8 27B (Xhigh)
  6. $1.031.03Step 5 Preview
  7. $1.131.13Mistral Large 4 Preview
  8. $1.241.24Gemini 3.8 Flash (High)
  9. $1.991.99Gemini 4 Argon (High)
  10. $2.012.01GLM-5.3 (Max)
Data from Artificial Analysis

Time per task

seconds per task · #5 of 26
  1. 5555K2 Horizon 375B A23B
  2. 5757Beam 1.0
  3. 9696Nemotron 3.5 Super VL
  4. 165165Inkling (Xhigh)
  5. 263263Step 5 Preview
  6. 284284Muse Spark Open (Xhigh)
  7. 341341Muse Wonder (Max)
  8. 364364GPT-6 Luna (Max)
  9. 396396DeepSeek V4.1 Flash 1017 (Max)
  10. 412412GPT-6 Astra (Max)
Data from Artificial Analysis

Tokens per task

output tokens per task · #13 of 27
  1. 51.5K52KK2 Horizon 375B A23B
  2. 53.8K54KMuse Wonder Full (Max)
  3. 54.5K55KMuse Wonder (Max)
  4. 55.2K55KKimi K3.1 Pro
  5. 61.6K62KGemini 4 Argon (High)
  6. 64.1K64KStep 5 Preview
  7. 64.3K64KMiMo-V2.6-Pro
  8. 66.8K67KQwen3.8 27B (Xhigh)
  9. 68.7K69KGLM-5.3-Flash
  10. 69.5K70KMistral Large 4 Preview
Data from Artificial Analysis

Speed and price

Speed

tokens per second · #17 of 162
  1. 241241NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)
  2. 234234Claude Haiku 5.5 (Max)
  3. 233233gpt-oss-120b (High)
  4. 232232Qwen3.5 Omni Flash
  5. 231231Nemotron 3 Nano Omni 30B A3B Reasoning
  6. 229229Step 5 Preview
  7. 223223Inkling Small
  8. 213213Ling 3.1 Flash
  9. 209209Ministral 3 3B
  10. 202202NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)
Data from Artificial Analysis

Latency

seconds to first token · #47 of 162
  1. 11Nova 2.0 Lite (Non-reasoning)
  2. 11NVIDIA Nemotron Nano 9B V2 (Non-reasoning)
  3. 1.11.1NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)
  4. 1.11.1Muse Glimmer (High)
  5. 1.11.1Gemma 4 31B (Reasoning)
  6. 1.21.2Step 5 Preview
  7. 1.21.2Nemotron 3 Ultra 550B A55B (Reasoning)
  8. 1.21.2DeepSeek V4.1 Flash 1017 (Max)
  9. 1.31.3Quasar 438B (Max, Based on GLM-5.2)
  10. 1.31.3Trinity Large Thinking
Data from Artificial Analysis

Price

$ per 1M tokens (3:1) · #107 of 169
  1. $1.131.13Nemotron 3 Ultra 550B A55B (Reasoning)
  2. $1.131.13Qwen3.8 27B (Non-reasoning)
  3. $1.31.3LongCat 2.0
  4. $1.351.35Qwen3.5 397B A17B (Non-reasoning)
  5. $1.351.35Qwen3.5 397B A17B (Reasoning)
  6. $1.431.43Step 5 Preview
  7. $1.51.5Gemini 3.8 Flash (High)
  8. $1.51.5Gemini 3.8 Flash (Medium)
  9. $1.51.5Gemini 3.8 Flash (Low)
  10. $1.51.5Qwen3.5 Omni Plus
Data from Artificial Analysis

Nearby models

  • GPT-6 Astra (Low)OpenAI45.8
  • Qwen3.8 Max (0902)Alibaba45.4
  • DeepSeek V4.1 Flash 1017 (Max)DeepSeek45.1
  • GLM-5.3 (Max)Z AI44.8
  • Ling 3.1 Flash 1019InclusionAI43.9
  • Claude Haiku 5.5 (Max)Anthropic43.4
  • MiniMax-M3.1-FlashMiniMax42.7
  • Claude Opus 5.5 (Low)Anthropic42.3
Data from Artificial Analysis

Same lab

  • Step3 VL 10BStepFun7.7
Data from Artificial Analysis
HorizonVascend IDEOlympusTalosCoursesCommunityBenchmark
DiscordGitHubXLet’s talkLegalPrivacyCookiesData useSecurity
© 2026 Vascend Technologies · VAT IT04911550616
ascend