Back to Models
Zhipu AIOpen Source

GLM 5.1

GLM 5.1 is a reasoning model developed by Zhipu AI. It has a large parameter count of approximately 754.0B and a context window of 200K, providing advanced reasoning capabilities.

Parameters

754.0B

Context Window

200K

License

MIT

Release Date

2026-03-27

API Pricing

Input Price (per 1M tokens)

$1.4

Output Price (per 1M tokens)

$

Billing Mode: standard

Strengths

  • Large 754.0B parameter scale
  • Long-context processing of 200K tokens
  • Open use via MIT license

Weaknesses

  • Extremely large 1.51TB file size
  • Requires vast compute resources to operate
  • Loading the model takes time

Use Cases

  • Reasoning tasks requiring complex logical thinking
  • Processing and analyzing vast documents
  • Foundational R&D for large language models

Deep Analysis

Parameters

754B total / 40B active MoE

Code Arena Elo

1530

#3 overall, first open-weight in top 3

SWE-Bench Pro

58.4%

#1 at launch

GPQA Diamond

86.2%

AIME 2026

95.3%

HLE (with tools)

52.3

Leads Opus 4.6 (45.0)

Pricing

$1.40/$4.40 per 1M tokens

License

MIT

Release Date

April 7, 2026

Strengths

  • First open-weight in Code Arena top 3 (1530 Elo)
  • SWE-Bench Pro 58.4% #1 at launch
  • Leads on MCP-Atlas, NL2Repo, CyberGym
  • MIT license
  • 8+ hour autonomous tasks

Weaknesses

  • GPQA 86.2 trails Opus 91.3
  • 200K context limiting for monorepos
  • 3x peak-hour pricing surcharge
  • 1.5TB FP8 weights need 8x H200

Competitor Comparison

ModelArenaSWEGPQAPrice
Claude Opus 4.61548 Code57.3 Pro91.3$5/$25
GPT-5.457.7 Pro92.0$2.50/$10
Kimi K2.658.6 Pro$0.60/$2.00

GLM-5.1 is a post-training refinement optimized for agentic engineering. First open-weight in Code Arena top 3 (1530 Elo), SWE-Bench Pro 58.4% #1 at launch.

Sources

Analysis generated: 2026-05-24