harn-vm 0.9.13

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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321

# ── Models ───────────────────────────────────────────────────────────────────
# Canonical model metadata: display name, provider, context window,
# capabilities, pricing (USD per 1M tokens), and deprecation status.
# Pricing reflects public provider pages snapshotted at the comment
# beside each section; edit the literal here and the change shows up in
# `git blame`.

# Anthropic ─ pricing pages: https://www.anthropic.com/pricing &
# https://platform.claude.com/docs/en/about-claude/model-deprecations.
# Verified 2026-07-02 against the provider /v1/models API and the
# deprecations page: Sonnet 4 / Opus 4 are retired, Opus 4.1 is deprecated
# (retires 2026-08-05), Opus 3 is retired, dated Sonnet 4.5 remains active,
# and Sonnet 5 joined the lineup. Since the 4.6 generation the full 1M-token
# context window ships at standard per-token pricing (no long-context beta
# or premium tier), so 4.6+ rows carry context_window = 1000000.

[models."claude-3-5-haiku-20241022"]
name = "Claude Haiku 3.5"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 0.80, output_per_mtok = 4.00, cache_read_per_mtok = 0.08, cache_write_per_mtok = 1.00 }
deprecated = true
deprecation_note = "Retired 2026-02-19 per Anthropic deprecations page. Use claude-haiku-4-5-20251001."
superseded_by = "claude-haiku-4-5-20251001"
tier = "small"
open_weight = false
strengths = ["speed", "cheap", "summarization", "tool_use"]

[models."claude-haiku-4-5-20251001"]
name = "Claude Haiku 4.5"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 1.00, output_per_mtok = 5.00, cache_read_per_mtok = 0.10, cache_write_per_mtok = 1.25 }
tier = "mid"
open_weight = false
strengths = ["speed", "cheap", "coding", "tool_use", "summarization"]

[models."claude-3-5-sonnet-20240620"]
name = "Claude Sonnet 3.5 (2024-06-20)"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 3.00, output_per_mtok = 15.00, cache_read_per_mtok = 0.30, cache_write_per_mtok = 3.75 }
deprecated = true
deprecation_note = "Retired 2025-10-28 per Anthropic deprecations page. Use claude-sonnet-4-5-20250929 or claude-sonnet-4-6."
superseded_by = "claude-sonnet-4-6"
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context"]
[models."claude-3-5-sonnet-20241022"]
name = "Claude Sonnet 3.5 (2024-10-22)"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 3.00, output_per_mtok = 15.00, cache_read_per_mtok = 0.30, cache_write_per_mtok = 3.75 }
deprecated = true
deprecation_note = "Retired 2025-10-28 per Anthropic deprecations page. Use claude-sonnet-4-5-20250929 or claude-sonnet-4-6."
superseded_by = "claude-sonnet-4-6"
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context"]
[models."claude-sonnet-4-20250514"]
name = "Claude Sonnet 4"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 3.00, output_per_mtok = 15.00, cache_read_per_mtok = 0.30, cache_write_per_mtok = 3.75 }
deprecated = true
deprecation_note = "Sunset 2026-06-15 per Anthropic deprecations page. Replaced by claude-sonnet-4-6."
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 49.0 }
[models."claude-sonnet-4-5"]
name = "Claude Sonnet 4.5"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 3.00, output_per_mtok = 15.00, cache_read_per_mtok = 0.30, cache_write_per_mtok = 3.75 }
deprecated = true
deprecation_note = "Undated compatibility selector. Anthropic's active API id is claude-sonnet-4-5-20250929; prefer that dated id or claude-sonnet-4-6."
superseded_by = "claude-sonnet-4-5-20250929"
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 77.2 }
[models."claude-sonnet-4-5-20250929"]
name = "Claude Sonnet 4.5"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 3.00, output_per_mtok = 15.00, cache_read_per_mtok = 0.30, cache_write_per_mtok = 3.75 }
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 77.2 }
[models."claude-sonnet-4-6"]
name = "Claude Sonnet 4.6"
provider = "anthropic"
context_window = 1000000
equivalence_group = "frontier-agent-coding"
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 3.00, output_per_mtok = 15.00, cache_read_per_mtok = 0.30, cache_write_per_mtok = 3.75 }
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 79.6 }
complementary_with = ["openai-gpt", "google-gemini", "qwen", "deepseek", "kimi"]
# NOTE: an erroneous `claude-sonnet-4-7` row used to live here. No such
# model exists — Anthropic's models overview (verified 2026-07-02) goes
# Sonnet 4.5 -> 4.6 -> 5; only the Opus line had a 4.7 release.

# Claude Sonnet 5 (released 2026-05) — the current Sonnet generation.
# Same request surface as Opus 4.8: adaptive thinking only (extended
# thinking budgets return 400), `output_config.effort` defaults to high,
# sampling params rejected, no assistant prefill. 1M-token context at
# standard pricing. Introductory pricing ($2 / $10 per MTok in/out,
# cache read 0.1x, 5-min cache write 1.25x) runs through 2026-08-31;
# standard $3 / $15 applies from 2026-09-01 — update this row then.
[models."claude-sonnet-5"]
name = "Claude Sonnet 5"
provider = "anthropic"
context_window = 1000000
equivalence_group = "frontier-agent-coding"
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 2.00, output_per_mtok = 10.00, cache_read_per_mtok = 0.20, cache_write_per_mtok = 2.50 }
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context", "agentic"]
complementary_with = ["openai-gpt", "google-gemini", "qwen", "deepseek", "kimi"]

# OpenRouter-routed Anthropic models. Kept as distinct catalog entries
# so `openrouter:anthropic/claude-*` resolves with the right capability
# matrix — without these, native-tools requests fail with `option `tools`
# is not supported by ... (provider openrouter)` because the lookup
# falls back to a no-tools shape. OpenRouter passes Anthropic's
# native-tools API through verbatim, including prompt-caching headers
# (cache attribution surface differs — tracked separately in #2320).
# Pricing matches the direct Anthropic API; OpenRouter adds its own
# margin at request time.
[models."anthropic/claude-haiku-4-5"]
name = "Claude Haiku 4.5 (via OpenRouter)"
provider = "openrouter"
context_window = 200000
capabilities = ["tools", "vision", "streaming", "prompt_caching"]
pricing = { input_per_mtok = 1.00, output_per_mtok = 5.00, cache_read_per_mtok = 0.10, cache_write_per_mtok = 1.25 }
tier = "mid"
open_weight = false
strengths = ["speed", "cheap", "coding", "tool_use", "summarization"]
[models."anthropic/claude-fable-5"]
name = "Claude Fable 5 (via OpenRouter)"
provider = "openrouter"
context_window = 1000000
capabilities = ["tools", "vision", "streaming", "prompt_caching"]
pricing = { input_per_mtok = 10.00, output_per_mtok = 50.00, cache_read_per_mtok = 1.00, cache_write_per_mtok = 12.50 }
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_pro = 80.3 }
[models."anthropic/claude-sonnet-4-6"]
name = "Claude Sonnet 4.6 (via OpenRouter)"
provider = "openrouter"
context_window = 1000000
equivalence_group = "frontier-agent-coding"
capabilities = ["tools", "vision", "streaming", "prompt_caching"]
pricing = { input_per_mtok = 3.00, output_per_mtok = 15.00, cache_read_per_mtok = 0.30, cache_write_per_mtok = 3.75 }
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context", "agentic"]
complementary_with = ["openai-gpt", "google-gemini", "qwen", "deepseek", "kimi"]
[models."anthropic/claude-sonnet-5"]
name = "Claude Sonnet 5 (via OpenRouter)"
provider = "openrouter"
context_window = 1000000
equivalence_group = "frontier-agent-coding"
capabilities = ["tools", "vision", "streaming", "prompt_caching"]
pricing = { input_per_mtok = 2.00, output_per_mtok = 10.00, cache_read_per_mtok = 0.20, cache_write_per_mtok = 2.50 }
tier = "frontier"
open_weight = false
strengths = ["coding", "reasoning", "tool_use", "long_context", "agentic"]
complementary_with = ["openai-gpt", "google-gemini", "qwen", "deepseek", "kimi"]
[models."claude-3-opus-20240229"]
name = "Claude Opus 3"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 15.00, output_per_mtok = 75.00, cache_read_per_mtok = 1.50, cache_write_per_mtok = 18.75 }
deprecated = true
deprecation_note = "Retired 2026-01-05 per Anthropic deprecations page. Use claude-opus-4-8."
superseded_by = "claude-opus-4-8"
tier = "frontier"
open_weight = false
strengths = ["reasoning", "long_context"]
[models."claude-opus-4-20250514"]
name = "Claude Opus 4"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 15.00, output_per_mtok = 75.00, cache_read_per_mtok = 1.50, cache_write_per_mtok = 18.75 }
deprecated = true
deprecation_note = "Sunset 2026-06-15 per Anthropic deprecations page. Replaced by claude-opus-4-8."
superseded_by = "claude-opus-4-8"
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 77.6 }
[models."claude-opus-4-1-20250805"]
name = "Claude Opus 4.1"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 15.00, output_per_mtok = 75.00, cache_read_per_mtok = 1.50, cache_write_per_mtok = 18.75 }
deprecated = true
deprecation_note = "Deprecated 2026-06-05 per Anthropic deprecations page; retires 2026-08-05. Use claude-opus-4-8."
superseded_by = "claude-opus-4-8"
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 78.9 }
[models."claude-opus-4-5-20251101"]
name = "Claude Opus 4.5"
provider = "anthropic"
context_window = 200000
capabilities = ["tools", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 5.00, output_per_mtok = 25.00, cache_read_per_mtok = 0.50, cache_write_per_mtok = 6.25 }
superseded_by = "claude-opus-4-8"
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
[models."claude-opus-4-6"]
name = "Claude Opus 4.6"
provider = "anthropic"
context_window = 1000000
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 5.00, output_per_mtok = 25.00, cache_read_per_mtok = 0.50, cache_write_per_mtok = 6.25 }
deprecated = true
deprecation_note = "Superseded by claude-opus-4-8. No formal sunset yet; switch when convenient."
superseded_by = "claude-opus-4-8"
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 80.8, swe_bench_pro = 53.4 }
[models."claude-opus-4-7"]
name = "Claude Opus 4.7"
provider = "anthropic"
context_window = 1000000
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 5.00, output_per_mtok = 25.00, cache_read_per_mtok = 0.50, cache_write_per_mtok = 6.25 }
deprecated = true
deprecation_note = "Superseded by claude-opus-4-8. No formal sunset yet; switch when convenient."
superseded_by = "claude-opus-4-8"
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 87.6, swe_bench_pro = 64.3 }
# Fast mode (research preview): 6x standard pricing, ~2.5x output tok/s.
fast_mode = { param = "speed", value = "fast", beta_header = "fast-mode-2026-02-01", otps_speedup = 2.5, status = "research_preview", pricing = { input_per_mtok = 30.00, output_per_mtok = 150.00, cache_read_per_mtok = 3.00, cache_write_per_mtok = 37.50 }, note = "Claude API + Managed Agents only. Migrate to Opus 4.8 fast mode for the cheaper 2x rate." }

# Claude Opus 4.8 (released 2026-05-28) — Anthropic's most capable model
# for complex reasoning, long-horizon agentic coding, and high-autonomy
# work; the current `opus` alias target. Adaptive thinking only (extended
# thinking budgets return 400); the `effort` parameter controls reasoning
# depth and defaults to `high` on every surface. Sampling params
# (temperature/top_p/top_k) are rejected — steer via prompting instead.
# 1M-token context window at standard per-token pricing (the former
# long-context beta graduated; there is no premium tier), matching every
# 4.6+ Claude row. Pricing per the Anthropic models overview: $5 / $25
# per MTok in/out (cache read 0.1x, 5-min cache write 1.25x).
[models."claude-opus-4-8"]
name = "Claude Opus 4.8"
provider = "anthropic"
context_window = 1000000
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 5.00, output_per_mtok = 25.00, cache_read_per_mtok = 0.50, cache_write_per_mtok = 6.25 }
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 88.6, swe_bench_pro = 69.2 }
complementary_with = ["openai-gpt", "google-gemini", "qwen", "deepseek", "kimi"]
# Fast mode (research preview): 2x standard pricing, ~2.5x output tok/s.
fast_mode = { param = "speed", value = "fast", beta_header = "fast-mode-2026-02-01", otps_speedup = 2.5, status = "research_preview", pricing = { input_per_mtok = 10.00, output_per_mtok = 50.00, cache_read_per_mtok = 1.00, cache_write_per_mtok = 12.50 }, note = "Claude API + Managed Agents only (not Bedrock/Vertex/Foundry); excluded from Batch and Priority Tier. Switching speed invalidates the prompt cache. Waitlist/account-manager gated." }

[models."anthropic/claude-opus-4.8"]
name = "Claude Opus 4.8 (via OpenRouter)"
provider = "openrouter"
context_window = 1000000
capabilities = ["tools", "vision", "streaming", "prompt_caching"]
pricing = { input_per_mtok = 5.00, output_per_mtok = 25.00, cache_read_per_mtok = 0.50, cache_write_per_mtok = 6.25 }
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_verified = 88.6, swe_bench_pro = 69.2 }

# Claude Fable 5 (released 2026-06-09) — Anthropic's most capable widely
# released model: the first generally available Mythos-class model, a new
# tier *above* Opus (Opus 4.8 stays current and keeps the `opus` alias;
# nothing is superseded). Same request surface as Opus 4.7/4.8 — adaptive
# thinking only, sampling params rejected, no assistant prefill — plus one
# new constraint: an explicit `thinking: {type: "disabled"}` is also a 400
# (thinking is always on; omit the field instead, which our request
# builder already does for the Disabled config). Safety classifiers can
# decline requests with stop_reason="refusal"; the `fallbacks` beta
# retries on Opus 4.8. No fast mode tier. 1M-token context window at
# standard pricing, like every 4.6+ row. Pricing per the Anthropic
# pricing page:
# $10 / $50 per MTok in/out (cache read 0.1x, 5-min cache write 1.25x);
# Batch API 50% discount applies. SWE-bench Verified not published for
# Fable 5 itself at launch (only for Mythos Preview), so it is omitted.
[models."claude-fable-5"]
name = "Claude Fable 5"
provider = "anthropic"
context_window = 1000000
capabilities = ["tools", "vision", "streaming", "prompt_caching", "thinking"]
pricing = { input_per_mtok = 10.00, output_per_mtok = 50.00, cache_read_per_mtok = 1.00, cache_write_per_mtok = 12.50 }
tier = "frontier"
open_weight = false
strengths = ["reasoning", "coding", "long_context", "agentic"]
benchmarks = { swe_bench_pro = 80.3 }
complementary_with = ["openai-gpt", "google-gemini", "qwen", "deepseek", "kimi"]