harn-vm 0.9.8

Async bytecode virtual machine for the Harn programming language
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138

# Z.AI GLM family, served via the OpenAI-compatible `/api/paas/v4` endpoint
# (the `.../v1` base 404s). The live `/models` listing on 2026-07-03 reports:
# glm-4.5, glm-4.5-air, glm-4.6, glm-4.7, glm-5, glm-5-turbo, glm-5.1,
# glm-5.2. GLM-5.2 is the current 1M-context open-weight flagship; the GLM-4.x
# line stays catalogued as the cheap independent-bucket coding/judge routes
# (glm-4.6 is confirmed serving 200 at paas/v4). Pricing per docs.z.ai.

# GLM-4.6 — 200K-context coding model, confirmed serving at paas/v4.
[models."glm-4.6"]
name = "GLM 4.6"
provider = "zai"
context_window = 204800
capabilities = ["tools", "streaming", "thinking", "prompt_caching"]
pricing = { input_per_mtok = 0.60, output_per_mtok = 2.20, cache_read_per_mtok = 0.11 }
tier = "mid"
open_weight = true
strengths = ["coding", "agentic", "tool_use", "cheap"]

# GLM-4.5 — prior-generation coding flagship; GLM-4.5-Air is the lighter
# MoE variant. Kept as cheap independent-bucket routes.
[models."glm-4.5"]
name = "GLM 4.5"
provider = "zai"
context_window = 131072
capabilities = ["tools", "streaming", "thinking", "prompt_caching"]
pricing = { input_per_mtok = 0.60, output_per_mtok = 2.20, cache_read_per_mtok = 0.11 }
tier = "mid"
open_weight = true
strengths = ["coding", "agentic", "tool_use", "cheap"]
[models."glm-4.5-air"]
name = "GLM 4.5 Air"
provider = "zai"
context_window = 131072
capabilities = ["tools", "streaming", "thinking"]
pricing = { input_per_mtok = 0.20, output_per_mtok = 1.10, cache_read_per_mtok = 0.03 }
tier = "small"
open_weight = true
strengths = ["speed", "cheap", "coding", "tool_use"]

# GLM-4.7 — current GLM-4.x generation coding model.
[models."glm-4.7"]
name = "GLM 4.7"
provider = "zai"
context_window = 204800
capabilities = ["tools", "streaming", "thinking", "prompt_caching"]
pricing = { input_per_mtok = 0.60, output_per_mtok = 2.20, cache_read_per_mtok = 0.11 }
tier = "mid"
open_weight = true
strengths = ["coding", "agentic", "tool_use", "cheap"]

# GLM-5 family. GLM-5.2 is the current 1M-context open-weight flagship;
# GLM-5.1 / GLM-5 are kept for pinned callers. glm-5-turbo is the current
# accelerated tier. OpenRouter mirrors live below.
[models."glm-5-turbo"]
name = "GLM 5 Turbo"
provider = "zai"
context_window = 202752
capabilities = ["tools", "streaming", "thinking"]
pricing = { input_per_mtok = 0.60, output_per_mtok = 2.00, cache_read_per_mtok = 0.11 }
tier = "mid"
open_weight = true
strengths = ["speed", "coding", "agentic", "tool_use", "cheap"]
[models."glm-5"]
name = "GLM 5"
provider = "zai"
context_window = 202752
capabilities = ["tools", "streaming", "thinking", "prompt_caching"]
# docs.z.ai pricing verified 2026-07-02: $1.00 / $3.20 per MTok (cache
# hit $0.20). The prior 0.98/3.08 figures matched no published tier.
pricing = { input_per_mtok = 1.00, output_per_mtok = 3.20, cache_read_per_mtok = 0.20 }
tier = "frontier"
open_weight = true
strengths = ["coding", "agentic", "tool_use"]

[models."glm-5.1"]
name = "GLM 5.1"
provider = "zai"
context_window = 202752
capabilities = ["tools", "streaming", "thinking", "prompt_caching"]
pricing = { input_per_mtok = 1.40, output_per_mtok = 4.40, cache_read_per_mtok = 0.26 }
tier = "frontier"
open_weight = true
strengths = ["coding", "agentic", "tool_use", "reasoning", "long_context"]
benchmarks = { swe_bench_pro_lead = 1.0 }
[models."glm-5.2"]
name = "GLM 5.2"
provider = "zai"
context_window = 1048576
capabilities = ["tools", "streaming", "thinking", "prompt_caching"]
pricing = { input_per_mtok = 1.40, output_per_mtok = 4.40, cache_read_per_mtok = 0.26 }
tier = "frontier"
open_weight = true
strengths = ["coding", "agentic", "tool_use", "reasoning", "long_context"]
benchmarks = { swe_bench_pro_lead = 1.0 }

# OpenRouter mirror of GLM-5 family so callers without a Z.AI key still
# resolve a route. OR doesn't list GLM-4.6/4.7 — the canonical OR slugs
# are the GLM-5 generation.
[models."z-ai/glm-5"]
name = "GLM 5 (via OpenRouter)"
provider = "openrouter"
context_window = 202752
capabilities = ["tools", "streaming"]
pricing = { input_per_mtok = 1.20, output_per_mtok = 4.00 }
tier = "frontier"
open_weight = true
strengths = ["coding", "agentic"]
[models."z-ai/glm-5.1"]
name = "GLM 5.1 (via OpenRouter)"
provider = "openrouter"
context_window = 202752
capabilities = ["tools", "streaming"]
pricing = { input_per_mtok = 0.98, output_per_mtok = 3.08 }
tier = "frontier"
open_weight = true
strengths = ["coding", "agentic", "tool_use", "reasoning", "long_context"]
[models."z-ai/glm-5.2"]
name = "GLM 5.2 (via OpenRouter)"
provider = "openrouter"
context_window = 1048576
capabilities = ["tools", "streaming", "thinking", "prompt_caching"]
# OpenRouter `/api/v1/models` live snapshot on 2026-07-02 reports
# $0.93/$3.00 per MTok and native `reasoning_effort` support. A 2026-07-03
# refresh reports input_cache_read at $0.18/MTok.
pricing = { input_per_mtok = 0.93, output_per_mtok = 3.00, cache_read_per_mtok = 0.18 }
tier = "frontier"
open_weight = true
strengths = ["coding", "agentic", "tool_use", "reasoning", "long_context"]
[models."z-ai/glm-5v-turbo"]
name = "GLM 5V Turbo (via OpenRouter)"
provider = "openrouter"
context_window = 202752
capabilities = ["tools", "streaming", "vision"]
pricing = { input_per_mtok = 1.20, output_per_mtok = 4.00 }
# Inline metadata keeps this row independent of fragment-order defaults.
tier = "frontier"
open_weight = true
strengths = ["vision", "coding", "agentic"]