Snowflake CEO finds GLM-5.2 competitive with Opus 4.7 at a fraction of the cost

2026-06-24 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

Snowflake's CEO, Sridhar Ramaswamy, conducted a real-world programming benchmark comparing China's GLM-5.2 AI model with Anthropic's Opus 4.7. The test involved 103 coding tasks, run three times, requiring code compatible with both DuckDB and Snowflake. Results showed GLM-5.2 solved 66% of tasks, nearly matching Opus 4.7's 67%. While Opus 4.7 demonstrated higher first-attempt accuracy (53.7% vs. 47.6%) and better efficiency, requiring 80 iterations and 439 million tokens compared to GLM's 99 iterations and 860 million tokens, GLM-5.2 presents a significant cost advantage. GLM-5.2 is priced at \$4.40 per million output tokens, substantially cheaper than Opus 4.7's \$25.00 and GPT-5.5's \$30.00, creating considerable price pressure on Western AI companies.

Key takeaway

For Directors of AI/ML evaluating large language models for coding tasks, you should critically assess the total cost of ownership beyond raw performance metrics. While Western models like Opus 4.7 offer higher first-attempt accuracy, GLM-5.2's significantly lower token pricing, at \$4.40 per million output tokens, can offset its higher token consumption. Your team could achieve comparable task completion rates at a substantially reduced operational cost, potentially impacting your budget allocation for AI infrastructure.

Key insights

Chinese AI model GLM-5.2 offers competitive coding performance at a fraction of Western models' cost, intensifying market price pressure.

Principles

• High iteration counts do not guarantee correctness.
• Price-performance ratios can disrupt market valuations.
• Cross-platform validation is a key model strength.

Method

Snowflake's benchmark involved 103 coding tasks, each run three times, requiring models to generate code functional on both DuckDB and Snowflake platforms.

In practice

• Evaluate models on real-world, multi-platform coding tasks.
• Consider total cost of ownership, including token usage.
• Explore non-Western models for cost-effective solutions.

Topics

• Large Language Models
• AI Benchmarking
• GLM-5.2
• Claude Opus 4.7
• AI Pricing
• Code Generation

Best for: CTO, Machine Learning Engineer, Entrepreneur, AI Engineer, Director of AI/ML, Investor

Related on AIssential

• Renewed Claude for Opus 4.6. Cancelled it After GLM-5, Again… — GLM-5 offers Claude-level coding performance at 6% of the price, potentially ending $20/month AI subscriptions.
• GLM 5.2 Z.ai Coding Model. From Beginner to Power User. Locally & API & Benchmark — GLM 5.2 offers a cost-effective, high-performance coding model comparable to leading LLMs like Fable 5.
• On GLM 5.2 — GLM-5.2 reportedly achieves benchmark parity with leading US models, especially for coding tasks.
• TAI #210: GLM-5.2 Closes Most of the Open-Weight Gap in Ten Weeks — GLM-5.2 demonstrates rapid open-weight model advancement through architectural, long-context, and reinforcement learning innovations, closing the gap with frontier models.
• Open source llm (glm 4.7) matching closed models on coding benchmarks. Tested via api on real projects. — Open-source models are achieving competitive, specialized performance in domains like coding at significantly lower costs.
• 😺 LIVE NOW: GPT 5.5 (The Spud Model??) Just Dropped. Let's Break It. — GPT-5.5 is a new "worker model" from OpenAI, excelling in knowledge work benchmarks with competitive pricing.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.