GLM 5.2 vs Claude Fable vs Kimi 2.7 vs Kimi K2.7

Full comparison across coding benchmarks, pricing, speed, and features — June 2026

Quick Winners

Best Benchmark Score

Claude Fable

Leads on HumanEval (+2.2%), SWE-bench (+3.9%), and MMLU

Best Price

GLM 5.2

Output at $0.28/M — 53× cheaper than Fable, comparable to Kimi

Best Speed

Kimi 2.7

~90 tok/s edges out GLM 5.2 at ~85 tok/s

Best for Local Use

GLM 5.2

Only top-tier coding model with open weights + Ollama support

Best Value Overall

GLM 5.2

92% of Fable's coding performance at 2% of the cost

Feature	GLM 5.2	Claude Fable	Kimi 2.7	Kimi K2.7
Developer	Zhipu AI	Anthropic	Moonshot AI	Moonshot AI
Context Window	128K	200K	128K	128K
HumanEval	92.1%	94.3%	90.8%	91.5%
LiveCodeBench	68.4%	71.2%	66.9%	69.1%
SWE-bench	51.2%	55.1%	49.7%	53.8%
Input price (per 1M tok)	$0.14	$3.00	$0.12	$0.15
Output price (per 1M tok)	$0.28	$15.00	$0.30	$0.60
Speed (tokens/sec)	~85	~70	~90	~75
Open weights	✅ HuggingFace	❌	❌	❌
Coding Plan feature	✅	✅ (extended thinking)	✅	✅
OpenRouter access	✅	✅	✅	✅
Ollama / local run	✅	❌	❌	❌

We gave all three models the same prompt: "Build a REST API in Python with FastAPI that handles CRUD for a todo list with SQLite."

GLM 5.2

Generated a Coding Plan first (5 files), then complete working code with tests. Minor import issue in test file.

4.2/5

Claude Fable

Extended thinking produced clean architecture. Most complete output with error handling and migration scripts.

4.6/5

Kimi 2.7

Fast output, working code, but no tests and minimal error handling. Good for quick prototypes.

3.9/5

Choose GLM 5.2 if: you need a cheap coding API, want local deployment, or are running high-volume agentic tasks
Choose Claude Fable if: you need the absolute best quality and budget is not a constraint
Choose Kimi 2.7 if: speed is critical and you want a Chinese-developed alternative to GLM