TPSTokens Per Second
NewsModelsIDEsRankingsBenchmarksSecurity

Navigate

NewsModelsIDEsRankingsChangelogBenchmarksSecurityModel QuizWhat I MissedHistorySaved

Companies

AnthropicOpenAIDeepMindMeta AIDeepSeekxAIMistral AIPerplexity AI

StepFun

https://www.stepfun.com →

News

No articles yet.

Models

Step-3.5-Flash-Base

StepFun

active
Context262K
Input/1M$0.1

Mar 2, 2026

Step-3.5-Flash

StepFun

active

196B total / 11B active open-source MoE under Apache 2.0 with 256K context and high efficiency.

Context262K
Input/1M$0.1

Feb 1, 2026

Open weights

Step-2

StepFun

deprecated

StepFun's trillion-parameter MoE, ranked 1st among Chinese models on LiveBench at launch.

Context16K

Nov 19, 2024

Top Benchmark Scores

Full leaderboard →

GPQA

Step-3.5-Flash-Base
41.7%

HumanEval

Step-3.5-Flash-Base
81.1%

MATH

Step-3.5-Flash-Base
66.8%

MMLU

Step-3.5-Flash-Base
85.8%

MMLU-Pro

Step-3.5-Flash-Base
62.3%

Speed (tok/s)

Step-3.5-Flash-Base
300 tokens_per_sec

SWE-bench Verified

Step-3.5-Flash-Base
74.4%
TPS

Tokens Per Second. The fastest LLM news on the internet — tracked automatically every 15 minutes.

Coverage

  • Latest News
  • Model Database
  • AI IDEs
  • Compare IDEs
  • Changelog
  • Benchmarks
  • Compare Models

Companies

  • Anthropic
  • OpenAI
  • Google DeepMind
  • Meta AI
  • DeepSeek
  • xAI
  • Mistral AI
  • Perplexity AI

Guides

  • All Rankings
  • Best by Use Case
  • Best Coding LLM
  • Best Cheap LLM
  • Best Reasoning LLM
  • Best AI IDE
  • Compare Models
  • Compare IDEs
  • About TPS
  • RSS Feed
  • Atom Feed

© 2026 TPS — Tokens Per Second.

The fastest LLM news. All signal, no noise.