omnillm 0.1.5

Production-grade LLM API gateway with multi-key load balancing, per-key rate limiting, circuit breaking, and cost tracking
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

//! Model pricing and cost estimation.

use crate::budget::tracker::{usd_to_micro, MicroDollar};
use crate::types::{PromptCacheUsage, TokenUsage};

/// Pricing for a specific model, in micro-dollars per 1k tokens.
struct ModelPricing {
    /// Cost per 1k input tokens in micro-dollars.
    input_per_1k: MicroDollar,
    /// Cost per 1k cached OpenAI-style input tokens in micro-dollars.
    cached_input_per_1k: Option<MicroDollar>,
    /// Cost per 1k Claude-style cache hit/read tokens in micro-dollars.
    cache_read_per_1k: Option<MicroDollar>,
    /// Cost per 1k short-lived Claude-style cache write tokens in micro-dollars.
    cache_write_short_per_1k: Option<MicroDollar>,
    /// Cost per 1k long-lived Claude-style cache write tokens in micro-dollars.
    cache_write_long_per_1k: Option<MicroDollar>,
    /// Cost per 1k output tokens in micro-dollars.
    output_per_1k: MicroDollar,
}

impl ModelPricing {
    fn basic(input_per_1k: f64, output_per_1k: f64) -> Self {
        Self {
            input_per_1k: usd_to_micro(input_per_1k),
            cached_input_per_1k: None,
            cache_read_per_1k: None,
            cache_write_short_per_1k: None,
            cache_write_long_per_1k: None,
            output_per_1k: usd_to_micro(output_per_1k),
        }
    }

    fn openai(input_per_1k: f64, cached_input_per_1k: f64, output_per_1k: f64) -> Self {
        Self {
            input_per_1k: usd_to_micro(input_per_1k),
            cached_input_per_1k: Some(usd_to_micro(cached_input_per_1k)),
            cache_read_per_1k: None,
            cache_write_short_per_1k: None,
            cache_write_long_per_1k: None,
            output_per_1k: usd_to_micro(output_per_1k),
        }
    }

    fn claude(
        input_per_1k: f64,
        cache_write_short_per_1k: f64,
        cache_write_long_per_1k: f64,
        cache_read_per_1k: f64,
        output_per_1k: f64,
    ) -> Self {
        Self {
            input_per_1k: usd_to_micro(input_per_1k),
            cached_input_per_1k: None,
            cache_read_per_1k: Some(usd_to_micro(cache_read_per_1k)),
            cache_write_short_per_1k: Some(usd_to_micro(cache_write_short_per_1k)),
            cache_write_long_per_1k: Some(usd_to_micro(cache_write_long_per_1k)),
            output_per_1k: usd_to_micro(output_per_1k),
        }
    }
}

/// Estimate cost from token count before the request is sent.
///
/// Uses total tokens with input pricing as a conservative upper bound.
pub(crate) fn estimate(tokens: u32, model: &str) -> MicroDollar {
    let p = pricing(model);
    (tokens as u64 * p.input_per_1k) / 1000
}

/// Compute actual cost from the response's usage report.
pub(crate) fn actual(usage: &TokenUsage, model: &str) -> MicroDollar {
    let p = pricing(model);
    let input = prompt_cost(usage, &p);
    let output = (usage.completion_tokens as u64 * p.output_per_1k) / 1000;
    input + output
}

fn prompt_cost(usage: &TokenUsage, pricing: &ModelPricing) -> MicroDollar {
    let Some(prompt_cache) = usage.prompt_cache.as_ref() else {
        return (usage.prompt_tokens as u64 * pricing.input_per_1k) / 1000;
    };

    openai_cached_prompt_cost(usage, prompt_cache, pricing)
        + claude_cache_extra_cost(prompt_cache, pricing)
}

fn openai_cached_prompt_cost(
    usage: &TokenUsage,
    prompt_cache: &PromptCacheUsage,
    pricing: &ModelPricing,
) -> MicroDollar {
    let Some(cached_rate) = pricing.cached_input_per_1k else {
        return (usage.prompt_tokens as u64 * pricing.input_per_1k) / 1000;
    };

    let cached_tokens = prompt_cache
        .cached_input_tokens
        .unwrap_or(0)
        .min(usage.prompt_tokens);
    let uncached_tokens = usage.prompt_tokens.saturating_sub(cached_tokens);

    ((uncached_tokens as u64 * pricing.input_per_1k) / 1000)
        + ((cached_tokens as u64 * cached_rate) / 1000)
}

fn claude_cache_extra_cost(prompt_cache: &PromptCacheUsage, pricing: &ModelPricing) -> MicroDollar {
    if pricing.cache_read_per_1k.is_none()
        && pricing.cache_write_short_per_1k.is_none()
        && pricing.cache_write_long_per_1k.is_none()
    {
        return 0;
    }

    let read = prompt_cache
        .cache_read_input_tokens
        .zip(pricing.cache_read_per_1k)
        .map(|(tokens, rate)| (tokens as u64 * rate) / 1000)
        .unwrap_or(0);

    let creation = if prompt_cache.cache_creation_short_input_tokens.is_some()
        || prompt_cache.cache_creation_long_input_tokens.is_some()
    {
        prompt_cache
            .cache_creation_short_input_tokens
            .zip(pricing.cache_write_short_per_1k)
            .map(|(tokens, rate)| (tokens as u64 * rate) / 1000)
            .unwrap_or(0)
            + prompt_cache
                .cache_creation_long_input_tokens
                .zip(pricing.cache_write_long_per_1k)
                .map(|(tokens, rate)| (tokens as u64 * rate) / 1000)
                .unwrap_or(0)
    } else {
        prompt_cache
            .cache_creation_input_tokens
            .zip(pricing.cache_write_short_per_1k)
            .map(|(tokens, rate)| (tokens as u64 * rate) / 1000)
            .unwrap_or(0)
    };

    read + creation
}

/// Look up pricing for a model. Unknown models are charged at GPT-4o rates
/// as a conservative default.
fn pricing(model: &str) -> ModelPricing {
    match model {
        m if m.starts_with("gpt-5.5") => ModelPricing::openai(0.005, 0.0005, 0.030),
        m if m.starts_with("gpt-5.4-mini") => ModelPricing::openai(0.000750, 0.000075, 0.004500),
        m if m.starts_with("gpt-5.4") => ModelPricing::openai(0.002500, 0.000250, 0.015000),
        m if m.starts_with("gpt-4o-mini") => ModelPricing::basic(0.000150, 0.000600),
        m if m.starts_with("gpt-4o") => ModelPricing::basic(0.005, 0.015),
        m if m.starts_with("claude-opus-4") => {
            ModelPricing::claude(0.005, 0.00625, 0.010, 0.00050, 0.025)
        }
        m if m.starts_with("claude-sonnet-4") || m.starts_with("claude-3-5-sonnet") => {
            ModelPricing::claude(0.003, 0.00375, 0.006, 0.00030, 0.015)
        }
        m if m.starts_with("claude-haiku-4") => {
            ModelPricing::claude(0.001, 0.00125, 0.002, 0.00010, 0.005)
        }
        m if m.starts_with("claude-3-haiku") => {
            ModelPricing::claude(0.00025, 0.00030, 0.00050, 0.00003, 0.00125)
        }
        _ => ModelPricing::basic(0.005, 0.015),
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn usage(
        prompt_tokens: u32,
        completion_tokens: u32,
        prompt_cache: PromptCacheUsage,
    ) -> TokenUsage {
        TokenUsage {
            prompt_tokens,
            completion_tokens,
            total_tokens: None,
            prompt_cache: Some(prompt_cache),
        }
    }

    #[test]
    fn estimate_does_not_assume_prompt_cache_discount() {
        assert_eq!(estimate(1_000, "gpt-5.4"), usd_to_micro(0.0025));
    }

    #[test]
    fn actual_uses_openai_cached_input_rate_when_available() {
        let usage = usage(
            1_000,
            100,
            PromptCacheUsage {
                cached_input_tokens: Some(400),
                ..Default::default()
            },
        );

        assert_eq!(actual(&usage, "gpt-5.4"), 3_100);
    }

    #[test]
    fn actual_uses_claude_cache_read_and_generic_write_rates() {
        let usage = usage(
            100,
            50,
            PromptCacheUsage {
                cache_read_input_tokens: Some(1_000),
                cache_creation_input_tokens: Some(2_000),
                ..Default::default()
            },
        );

        assert_eq!(actual(&usage, "claude-3-5-sonnet-20241022"), 8_850);
    }

    #[test]
    fn actual_uses_claude_specific_short_and_long_write_rates() {
        let usage = usage(
            0,
            0,
            PromptCacheUsage {
                cache_creation_short_input_tokens: Some(1_000),
                cache_creation_long_input_tokens: Some(1_000),
                ..Default::default()
            },
        );

        assert_eq!(actual(&usage, "claude-sonnet-4-5"), 9_750);
    }

    #[test]
    fn unknown_cache_rates_fall_back_to_uncached_prompt_cost() {
        let usage = usage(
            1_000,
            0,
            PromptCacheUsage {
                cached_input_tokens: Some(900),
                cache_read_input_tokens: Some(2_000),
                cache_creation_input_tokens: Some(2_000),
                ..Default::default()
            },
        );

        assert_eq!(actual(&usage, "unknown-model"), usd_to_micro(0.005));
    }
}