@digital4better/data 1.3.62 → 1.3.63

package/data/ai/provenance.md

@@ -0,0 +1,308 @@
+# Parameter Provenance
+
+Audit date: 2026-10-01. Counts are in billions; columns always distinguish active from total parameters. See the [methodology and limitations](./README.md).
+
+This is a dated provenance record, not a claim that proprietary architectures have been disclosed. It covers the closed-model parameter estimates in this revision and three new open-weight family proxies. Open-weight model cards are linked directly in `models.json`.
+
+## LifeArchitect
+
+The following 65 entries use values read from the public [Models Table](https://lifearchitect.ai/models-table/), not a ratio inferred from its total column. Dense active counts equal total counts. All remain marked as estimates. GPT-4-32k shares the GPT-4 Classic family estimate rather than a separate measurement.
+
+| Model | Public table row | Active | Total |
+| --- | --- | ---: | ---: |
+| `amazon-nova-premier` | Nova Premier | 470 | 470 |
+| `amazon-nova-pro` | Nova Pro | 90 | 90 |
+| `claude-3-opus` | Claude 3 Opus | 125 | 2500 |
+| `claude-3.5-sonnet` | Claude 3.5 Sonnet | 400 | 400 |
+| `claude-3.5-sonnet-v2` | Claude 3.5 Sonnet (new) | 400 | 400 |
+| `claude-3.7-sonnet` | Claude 3.7 Sonnet | 400 | 400 |
+| `claude-fable-5` | Claude Fable 5 | 150 | 10000 |
+| `claude-fable-5.1` | Claude Fable 5.1 | 150 | 10000 |
+| `claude-mythos-5` | Claude Mythos 5 | 150 | 10000 |
+| `claude-mythos-5.1` | Claude Mythos 5.1 | 150 | 10000 |
+| `claude-mythos-preview` | Claude Mythos Preview | 150 | 10000 |
+| `claude-opus-4` | Claude Opus 4 | 150 | 6000 |
+| `claude-opus-4.1` | Claude Opus 4.1 | 150 | 5000 |
+| `claude-opus-4.5` | Claude Opus 4.5 | 150 | 5000 |
+| `claude-opus-4.6` | Claude Opus 4.6 | 150 | 5000 |
+| `claude-opus-4.7` | Claude Opus 4.7 | 150 | 5000 |
+| `claude-opus-4.8` | Claude Opus 4.8 | 150 | 5000 |
+| `claude-opus-5` | Claude Opus 5 | 150 | 5000 |
+| `claude-opus-5.5` | Claude Opus 5.5 | 150 | 3000 |
+| `claude-sonnet-4.5` | Claude Sonnet 4.5 | 20 | 1000 |
+| `claude-sonnet-4.6` | Claude Sonnet 4.6 | 20 | 1000 |
+| `claude-sonnet-5` | Claude Sonnet 5 | 20 | 1000 |
+| `claude-sonnet-5.5` | Claude Sonnet 5.5 | 40 | 2000 |
+| `gemini-3-flash` | Gemini 3 Flash | 10 | 200 |
+| `gemini-3-pro` | Gemini 3 Pro | 150 | 3000 |
+| `gemini-3.1-pro` | Gemini 3.1 Pro | 150 | 3000 |
+| `gemini-3.5-flash` | Gemini 3.5 Flash | 25 | 500 |
+| `gemini-3.6-flash` | Gemini 3.6 Flash | 25 | 500 |
+| `gemini-3.7-flash` | Gemini 3.7 Flash | 25 | 500 |
+| `gemini-3.8-flash` | Gemini 3.8 Flash | 25 | 500 |
+| `gpt-35-turbo` | ChatGPT (gpt-3.5-turbo) | 20 | 20 |
+| `gpt-4` | GPT-4 Classic | 88 | 1760 |
+| `gpt-4-32k` | GPT-4 Classic | 88 | 1760 |
+| `gpt-4-turbo` | GPT-4 Turbo | 3.5 | 70 |
+| `gpt-4.1` | GPT-4.1 | 15 | 300 |
+| `gpt-4.5` | GPT-4.5 | 225 | 4500 |
+| `gpt-4o` | GPT-4o | 10 | 200 |
+| `gpt-4o-mini` | GPT-4o mini | 0.4 | 8 |
+| `gpt-5` | GPT-5 | 150 | 3000 |
+| `gpt-5.1` | GPT-5.1 | 150 | 3000 |
+| `gpt-5.2` | GPT-5.2 | 150 | 3000 |
+| `gpt-5.3-chat` | GPT-5.3 Instant | 15 | 300 |
+| `gpt-5.4` | GPT-5.4 | 150 | 3000 |
+| `gpt-5.5` | GPT-5.5 | 150 | 3000 |
+| `gpt-5.6-cyber` | GPT-5.6-Cyber | 150 | 5000 |
+| `gpt-5.6-sol` | GPT-5.6 Sol | 150 | 5000 |
+| `gpt-6-astra` | GPT-6 Astra | 250 | 10000 |
+| `gpt-6-sol` | GPT-6 Sol | 150 | 2500 |
+| `gpt-6.1-sol` | GPT-6.1 Sol | 150 | 2500 |
+| `grok-4` | Grok 4 | 150 | 3000 |
+| `grok-4.3` | Grok 4.3 | 25 | 500 |
+| `grok-4.6` | Grok 4.6 | 75 | 1500 |
+| `grok-4.7` | Grok 4.7 | 105 | 2100 |
+| `grok-code-fast-1` | grok-code-fast-1 | 40 | 800 |
+| `mai-thinking-1` | MAI-Thinking-1 | 35 | 1000 |
+| `mistral-large-2402` | Mistral Large | 300 | 300 |
+| `mistral-medium-3` | Mistral Medium 3 | 50 | 50 |
+| `muse-spark-1.3` | Muse Spark 1.3 | 50 | 500 |
+| `nova-pro` | Nova Pro | 90 | 90 |
+| `o1` | o1 | 10 | 200 |
+| `o1-preview` | o1-preview | 10 | 200 |
+| `o3` | o3 | 30 | 600 |
+| `o3-mini` | o3-mini | 70 | 70 |
+| `o4-mini` | o4-mini | 10 | 200 |
+| `text-davinci-003` | text-davinci-003 | 175 | 175 |
+
+## EcoLogits And Adaptations
+
+Credit: EcoLogits contributors, including Samuel Rince. Sources: [pinned model registry](https://github.com/mlco2/ecologits/blob/c739e8d61f6a12a85864adb698ab85623fe70f3c/ecologits/data/models.json), [proprietary-model estimate table](https://docs.google.com/spreadsheets/d/1XkPTkrGxpwWpIVIxpVvgRJuInSZsqbndTQbFGcHhdd0/edit#gid=803926269) and [methodology](https://ecologits.ai/latest/methodology/proprietary_models/).
+
+These values are retained conservatively where their numerical lineage matches EcoLogits and no independent, checkpoint-specific replacement was verified. A matching value alone does not prove copying; attribution is retained rather than claiming independent derivation. The source row and transformation below distinguish EcoLogits estimates from local extensions.
+
+A midpoint collapses upstream uncertainty: it is not a more accurate measurement. Historical rounding is preserved (for example, 47.5 becomes 48). A family proxy reuses a different model row and must not be represented as an EcoLogits measurement of that variant.
+
+| Catalog model | EcoLogits source row | Source active range | Source total range | Catalog active / total | Adaptation |
+| --- | --- | --- | --- | --- | --- |
+| `claude-3.5-haiku` | `claude-3-5-haiku` | 8B-28B | 8B-28B | 18 / 18 | Midpoint, retaining rounding |
+| `claude-haiku-4.5` | `clause-haiku-4-5` | 10-35B | 10-35B | 23 / 23 | Midpoint, retaining rounding |
+| `claude-sonnet-4` | `claude-sonnet-4` | 44B-132B | 440B | 88 / 440 | Midpoint, retaining rounding |
+| `gemini-2.0-flash` | `gemini-2.0-flash` | 44B-132B | 440B | 88 / 440 | Midpoint, retaining rounding |
+| `gemini-2.0-flash-lite` | `gemini-2.0-flash-lite` | 8B-28B | 8B-28B | 18 / 18 | Midpoint, retaining rounding |
+| `gemini-2.5-flash` | `gemini-2.5-flash` | 44B-132B | 440B | 88 / 440 | Midpoint, retaining rounding |
+| `gemini-2.5-flash-image` | `gemini-2.5-flash` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gemini-2.5-flash-lite` | `gemini-2.5-flash-lite` | 8B-28B | 8B-28B | 18 / 18 | Midpoint, retaining rounding |
+| `gemini-2.5-flash-live-api` | `gemini-2.5-flash` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gemini-2.5-flash-tts` | `gemini-2.5-flash` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gemini-2.5-pro` | `gemini-2.5-pro` | 200B-600B | 2000B | 400 / 2000 | Midpoint, retaining rounding |
+| `gemini-2.5-pro-tts` | `gemini-2.5-pro` | 200B-600B | 2000B | 400 / 2000 | Local family proxy |
+| `gemini-3-pro-image` | `gemini-3-pro` | 120B-360B | 1200B | 240 / 1200 | Local family proxy |
+| `gemini-3.1-flash-image` | `gemini-3-pro` | 120B-360B | 1200B | 240 / 1200 | Local family proxy |
+| `gemini-3.1-flash-lite` | `gemini-3.1-flash-lite` | 30B-105B | 30B-105B | 68 / 68 | Midpoint, retaining rounding |
+| `gemini-3.1-flash-lite-image` | `gemini-3-flash` | 30B-100B | 300B | 65 / 300 | Local family proxy |
+| `gemini-3.1-flash-live` | `gemini-3-flash` | 30B-100B | 300B | 65 / 300 | Local family proxy |
+| `gemini-3.1-flash-tts` | `gemini-3-flash` | 30B-100B | 300B | 65 / 300 | Local family proxy |
+| `gemini-3.5-flash-lite` | `gemini-3.1-flash-lite` | 30B-105B | 30B-105B | 68 / 68 | Local family proxy |
+| `gemini-3.5-live-translate` | `gemini-3-flash` | 30B-100B | 300B | 65 / 300 | Local family proxy |
+| `gpt-35-turbo-16k` | `gpt-3.5-turbo` | 20B-70B | 20B-70B | 45 / 45 | Local family proxy |
+| `gpt-4-turbo-vision` | `gpt-4-turbo` | 88B-264B | 880B | 140 / 880 | Local active scaling; see note below |
+| `gpt-4.1-mini` | `gpt-4.1-mini` | 40B-112B | 40B-112B | 76 / 76 | Midpoint, retaining rounding |
+| `gpt-4.1-nano` | `gpt-4.1-nano` | 10B-37B | 10B-37B | 24 / 24 | Midpoint, retaining rounding |
+| `gpt-4o-audio` | `gpt-4o` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gpt-4o-audio-preview` | `gpt-4o` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gpt-4o-mini-audio` | `gpt-4o-mini` | 8B-28B | 8B-28B | 18 / 18 | Local family proxy |
+| `gpt-4o-mini-audio-preview` | `gpt-4o-mini` | 8B-28B | 8B-28B | 18 / 18 | Local family proxy |
+| `gpt-4o-mini-realtime` | `gpt-4o-mini` | 8B-28B | 8B-28B | 18 / 18 | Local family proxy |
+| `gpt-4o-mini-realtime-audio` | `gpt-4o-mini` | 8B-28B | 8B-28B | 18 / 18 | Local family proxy |
+| `gpt-4o-mini-realtime-preview` | `gpt-4o-mini` | 8B-28B | 8B-28B | 18 / 18 | Local family proxy |
+| `gpt-4o-mini-transcribe` | `gpt-4o-mini` | 8B-28B | 8B-28B | 18 / 18 | Local family proxy |
+| `gpt-4o-mini-tts` | `gpt-4o-mini` | 8B-28B | 8B-28B | 18 / 18 | Local family proxy |
+| `gpt-4o-realtime-preview` | `gpt-4o` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gpt-4o-transcribe` | `gpt-4o` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gpt-4o-transcribe-diarize` | `gpt-4o` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+| `gpt-5-chat` | `gpt-5` | 30B-90B | 300B | 60 / 300 | Local family proxy |
+| `gpt-5-codex` | `gpt-5` | 30B-90B | 300B | 60 / 300 | Local family proxy |
+| `gpt-5-mini` | `gpt-5-mini` | 25B-70B | 25B-70B | 48 / 48 | Midpoint, retaining rounding |
+| `gpt-5-nano` | `gpt-5-nano` | 5B-18.5B | 5B-18.5B | 12 / 12 | Midpoint, retaining rounding |
+| `gpt-5-pro` | `gpt-5-pro` | 360B-1080B | 3600B | 720 / 3600 | Midpoint, retaining rounding |
+| `gpt-5.4-mini` | `gpt-5.4-mini` | 56B-158B | 56B-158B | 107 / 107 | Midpoint, retaining rounding |
+| `gpt-5.4-nano` | `gpt-5.4-nano` | 15B-58B | 15B-58B | 37 / 37 | Midpoint, retaining rounding |
+| `gpt-5.4-pro` | `gpt-5.4-pro` | 540-1800B | 5400B | 1170 / 5400 | Midpoint, retaining rounding |
+| `gpt-6-luna` | `gpt-6-luna` | 3.8B-11B | 38B | 7.4 / 38 | Midpoint, retaining rounding |
+| `o1-mini` | `o1-mini` | 8B-28B | 8B-28B | 18 / 18 | Midpoint, retaining rounding |
+| `o1-pro` | `o1` | 44B-132B | 440B | 88 / 440 | Local family proxy |
+
+`gpt-4-turbo-vision`: the retained 880 B total follows the EcoLogits GPT-4 Turbo row. The 140 B active count is a historical local scaling assumption (half the former 280 B GPT-4 estimate), not EcoLogits' 176 B midpoint. It remains unvalidated and is not recalculated from the newly selected GPT-4 estimate.
+
+The pinned EcoLogits repository is published under [MPL-2.0](https://github.com/mlco2/ecologits/blob/c739e8d61f6a12a85864adb698ab85623fe70f3c/LICENSE). These notices and links identify the reused estimates and our transformations; they do not replace the upstream license or imply that attribution alone resolves every reuse obligation. Upstream rights are preserved alongside this repository's [license](../../LICENSE).
+
+## Open-Weight Family Proxies
+
+| Model | Proxy source | Active | Total |
+| --- | --- | ---: | ---: |
+| `glm-4.7` | [GLM-4.5](https://huggingface.co/zai-org/GLM-4.5) | 32 | 355 |
+| `glm-5.2` | [GLM-5](https://huggingface.co/zai-org/GLM-5) | 40 | 744 |
+| `glm-5.3` | [GLM-5](https://huggingface.co/zai-org/GLM-5) | 40 | 744 |
+
+These three sizes are explicitly estimated; the linked parent checkpoint does not establish the parameter count of each later checkpoint. GLM-5.3-Flash has its own published 320 B / 18 B specification and does not use this proxy.
+
+## Local Hypotheses
+
+41 entries use explicitly local sizing assumptions. Undisclosed parameter/architecture values below have low confidence: they make calculations possible, not the proprietary weights known. The source links establish the anchors or public API facts, not the inferred target size. The original Ministral 3B has a published nominal total, but its dense active-count assumption is distinguished from that fact.
+
+Scenarios are compared before choosing a point estimate. A documented parent takes precedence over a token-price ratio. Otherwise the selected analogue or rounded geometric midpoint is stated per row. A scenario spread is a sensitivity check, not a statistical confidence interval; correlated estimates are not independent confirmations. No new hypotheses were created just to remove an EcoLogits attribution.
+
+Prices are the standard public rates consulted on 2026-10-01, except the Grok 4.1 Fast anchor from its 2025-11-19 release announcement. They exclude batch/cached discounts and may change independently of model size. TTS comparisons use characters, not tokens. See the [calculation rules](./README.md#calculation-rules).
+
+| Model | Selected active / total (B) | Scenarios, selection and limits | Numerical anchors |
+| --- | ---: | --- | --- |
+| `codex-mini` | 10 / 200 | OpenAI explicitly calls this a fine-tuned o4-mini. Select the LifeArchitect parent estimate 10 / 200 B. A price ratio $6 / $4.40 per million output tokens gives 13.64 / 272.73 B, 1.36x higher; rejected as the primary estimate because documented fine-tuning is stronger evidence than price. | [anchor 1](https://lifearchitect.ai/models-table/), [anchor 2](https://developers.openai.com/api/docs/models/o4-mini) |
+| `cohere-parse-v5` | 8 / 8 | Visual document-parser analogue: Cohere Aya Vision 8 B versus 32 B (4x spread). Select 8 B as a compact OCR-oriented scenario, not confirmed ancestry. No token context asserted: the documented API accepts document images rather than an exposed conversational token window. | [anchor 1](https://huggingface.co/CohereLabs/aya-vision-8b), [anchor 2](https://huggingface.co/CohereLabs/aya-vision-32b) |
+| `computer-use-preview` | 10 / 200 | Multimodal agent analogue: LifeArchitect GPT-4o 10 active / 200 total. Price scenario using official output prices $12 vs $10 per million text tokens gives 12 / 240 B (1.2x). Prefer the unscaled analogue: tool specialization does not establish larger weights. No confirmed parentage. | [anchor 1](https://lifearchitect.ai/models-table/), [anchor 2](https://developers.openai.com/api/docs/models/gpt-4o) |
+| `flux-1-kontext-pro` | 12 / 12 | Same-generation open-family proxy: 12 B dense and hidden width 3072. FLUX.1 (12 B) versus FLUX.2 (32 B) gives a 2.67x cross-generation alternative, but the matching generation is preferred. Pro/flex labels and inference steps alone do not establish weight counts. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://github.com/black-forest-labs/flux/blob/main/src/flux/util.py) |
+| `flux-1.1-pro` | 12 / 12 | Same-generation open-family proxy: 12 B dense and hidden width 3072. FLUX.1 (12 B) versus FLUX.2 (32 B) gives a 2.67x cross-generation alternative, but the matching generation is preferred. Pro/flex labels and inference steps alone do not establish weight counts. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://github.com/black-forest-labs/flux/blob/main/src/flux/util.py) |
+| `flux-2-flex` | 32 / 32 | Same-generation open-family proxy: 32 B dense and hidden width 6144. FLUX.1 (12 B) versus FLUX.2 (32 B) gives a 2.67x cross-generation alternative, but the matching generation is preferred. Pro/flex labels and inference steps alone do not establish weight counts. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 2](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py) |
+| `flux-2-pro` | 32 / 32 | Same-generation open-family proxy: 32 B dense and hidden width 6144. FLUX.1 (12 B) versus FLUX.2 (32 B) gives a 2.67x cross-generation alternative, but the matching generation is preferred. Pro/flex labels and inference steps alone do not establish weight counts. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 2](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py) |
+| `gemini-3.8-flash-cyber` | 25 / 500 | LifeArchitect Gemini-3.8-Flash family-scale proxy 25 / 500 B; compare Gemini-3-Flash 10 / 200 B (2.5x spread). Retain the more recent anchor. Google documents post-training of 3.8 Flash, supporting a parent-weight proxy. | [anchor 1](https://lifearchitect.ai/models-table/) |
+| `gemini-3.8-live` | 25 / 500 | LifeArchitect Gemini-3.8-Flash family-scale proxy 25 / 500 B; compare Gemini-3-Flash 10 / 200 B (2.5x spread). Retain the more recent anchor. Naming/capability analogy is not evidence of shared weights, especially for generative video or live audio components. | [anchor 1](https://lifearchitect.ai/models-table/) |
+| `gemini-embedding-001` | 3.1 / 3.1 | Dense text-embedding scenarios: Gecko 1.2 B and Qwen3-Embedding 8 B (6.67x spread); sqrt(1.2 * 8) = 3.098, rounded to 3.1 B. Cross-family analogy, not a Gemini architecture disclosure. Output dimension cannot determine model size. | [anchor 1](https://arxiv.org/abs/2403.20327), [anchor 2](https://huggingface.co/Qwen/Qwen3-Embedding-8B) |
+| `gemini-omni-1.1-flash` | 25 / 500 | LifeArchitect Gemini-3.8-Flash family-scale proxy 25 / 500 B; compare Gemini-3-Flash 10 / 200 B (2.5x spread). Retain the more recent anchor. Naming/capability analogy is not evidence of shared weights, especially for generative video or live audio components. | [anchor 1](https://lifearchitect.ai/models-table/) |
+| `gemini-omni-flash` | 25 / 500 | LifeArchitect Gemini-3.8-Flash family-scale proxy 25 / 500 B; compare Gemini-3-Flash 10 / 200 B (2.5x spread). Retain the more recent anchor. Naming/capability analogy is not evidence of shared weights, especially for generative video or live audio components. | [anchor 1](https://lifearchitect.ai/models-table/) |
+| `gpt-image-2.5-flare` | 20 / 20 | Open image-generator analogues: FLUX.1-dev 12 B and FLUX.2-dev 32 B; sqrt(12 * 32) = 19.60, rounded to 20 B dense (2.67x spread). This estimates the core generator, not all service components. No size multiplier for Flash/Pro or quality presets. Hidden width 6144 is borrowed from FLUX.2, not measured. Context 32000 is an image-API analogue from Azure FLUX.2, not an OpenAI disclosure. Flare and Sunburst share $30/M image-output-token pricing, which does not prove equal weights. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 3](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py), [anchor 4](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure) |
+| `gpt-image-2.5-sunburst` | 20 / 20 | Open image-generator analogues: FLUX.1-dev 12 B and FLUX.2-dev 32 B; sqrt(12 * 32) = 19.60, rounded to 20 B dense (2.67x spread). This estimates the core generator, not all service components. No size multiplier for Flash/Pro or quality presets. Hidden width 6144 is borrowed from FLUX.2, not measured. Context 32000 is an image-API analogue from Azure FLUX.2, not an OpenAI disclosure. Flare and Sunburst share $30/M image-output-token pricing, which does not prove equal weights. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 3](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py), [anchor 4](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure) |
+| `grok-4.1-fast-non-reasoning` | 25 / 500 | Price scenario: Grok-4.3 LifeArchitect 25 / 500 B scaled by $0.50 / $2.50 per million output tokens = 5 / 100 B. Family-scale scenario: retain 25 / 500 B (5x disagreement). Select the unscaled 25 / 500 B, because a Fast endpoint can reduce cost without proportional weight reduction. This is NOT LifeArchitect's 4.1-Fast estimate. No architecture/count difference assumed between reasoning modes. | [anchor 1](https://lifearchitect.ai/models-table/), [anchor 2](https://docs.x.ai/developers/models/grok-4.3), [anchor 3](https://x.ai/news/grok-4-1-fast) |
+| `grok-4.1-fast-reasoning` | 25 / 500 | Price scenario: Grok-4.3 LifeArchitect 25 / 500 B scaled by $0.50 / $2.50 per million output tokens = 5 / 100 B. Family-scale scenario: retain 25 / 500 B (5x disagreement). Select the unscaled 25 / 500 B, because a Fast endpoint can reduce cost without proportional weight reduction. This is NOT LifeArchitect's 4.1-Fast estimate. No architecture/count difference assumed between reasoning modes. | [anchor 1](https://lifearchitect.ai/models-table/), [anchor 2](https://docs.x.ai/developers/models/grok-4.3), [anchor 3](https://x.ai/news/grok-4-1-fast) |
+| `grok-4.20-non-reasoning` | 25 / 500 | Grok-4.3 analogue: LifeArchitect 25 / 500 B. Same-provider standard output prices $2.50 / $2.50 imply ratio 1 and the same 25 / 500 B scenario. Shared sizing anchor means these are not independent confirmations. MoE and the 5% active fraction remain hypotheses. | [anchor 1](https://lifearchitect.ai/models-table/), [anchor 2](https://docs.x.ai/developers/models/grok-4.3), [anchor 3](https://x.ai/news/grok-4-1-fast) |
+| `grok-4.20-reasoning` | 25 / 500 | Grok-4.3 analogue: LifeArchitect 25 / 500 B. Same-provider standard output prices $2.50 / $2.50 imply ratio 1 and the same 25 / 500 B scenario. Shared sizing anchor means these are not independent confirmations. MoE and the 5% active fraction remain hypotheses. | [anchor 1](https://lifearchitect.ai/models-table/), [anchor 2](https://docs.x.ai/developers/models/grok-4.3), [anchor 3](https://x.ai/news/grok-4-1-fast) |
+| `lyria-002` | 3.3 / 3.3 | Open music-generator scenarios: MusicGen medium 1.5 B and large 3.3 B (2.2x spread). Select 3.3 B dense as the larger analogue; image conditioning, singing and proprietary codecs may add uncounted weights. Clip duration and Pro naming do not determine parameter counts. | [anchor 1](https://huggingface.co/facebook/musicgen-medium), [anchor 2](https://huggingface.co/facebook/musicgen-large) |
+| `lyria-3-clip` | 3.3 / 3.3 | Open music-generator scenarios: MusicGen medium 1.5 B and large 3.3 B (2.2x spread). Select 3.3 B dense as the larger analogue; image conditioning, singing and proprietary codecs may add uncounted weights. Clip duration and Pro naming do not determine parameter counts. | [anchor 1](https://huggingface.co/facebook/musicgen-medium), [anchor 2](https://huggingface.co/facebook/musicgen-large) |
+| `lyria-3-pro` | 3.3 / 3.3 | Open music-generator scenarios: MusicGen medium 1.5 B and large 3.3 B (2.2x spread). Select 3.3 B dense as the larger analogue; image conditioning, singing and proprietary codecs may add uncounted weights. Clip duration and Pro naming do not determine parameter counts. | [anchor 1](https://huggingface.co/facebook/musicgen-medium), [anchor 2](https://huggingface.co/facebook/musicgen-large) |
+| `mai-image-2.5` | 20 / 20 | Open image-generator analogues: FLUX.1-dev 12 B and FLUX.2-dev 32 B; sqrt(12 * 32) = 19.60, rounded to 20 B dense (2.67x spread). This estimates the core generator, not all service components. No size multiplier for Flash/Pro or quality presets. Hidden width 6144 is borrowed from FLUX.2, not measured. Context 32000 is published by Azure. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 3](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py) |
+| `mai-image-2.5-flash` | 20 / 20 | Open image-generator analogues: FLUX.1-dev 12 B and FLUX.2-dev 32 B; sqrt(12 * 32) = 19.60, rounded to 20 B dense (2.67x spread). This estimates the core generator, not all service components. No size multiplier for Flash/Pro or quality presets. Hidden width 6144 is borrowed from FLUX.2, not measured. Context 32000 is published by Azure. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 3](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py) |
+| `mai-image-2.5-pro` | 20 / 20 | Open image-generator analogues: FLUX.1-dev 12 B and FLUX.2-dev 32 B; sqrt(12 * 32) = 19.60, rounded to 20 B dense (2.67x spread). This estimates the core generator, not all service components. No size multiplier for Flash/Pro or quality presets. Hidden width 6144 is borrowed from FLUX.2, not measured. Context 32000 is published by Azure. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 3](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py) |
+| `mai-image-2.6` | 20 / 20 | Open image-generator analogues: FLUX.1-dev 12 B and FLUX.2-dev 32 B; sqrt(12 * 32) = 19.60, rounded to 20 B dense (2.67x spread). This estimates the core generator, not all service components. No size multiplier for Flash/Pro or quality presets. Hidden width 6144 is borrowed from FLUX.2, not measured. Context 32000 is published by Azure. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 3](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py) |
+| `mai-image-2.6-flash` | 20 / 20 | Open image-generator analogues: FLUX.1-dev 12 B and FLUX.2-dev 32 B; sqrt(12 * 32) = 19.60, rounded to 20 B dense (2.67x spread). This estimates the core generator, not all service components. No size multiplier for Flash/Pro or quality presets. Hidden width 6144 is borrowed from FLUX.2, not measured. Context 32000 is published by Azure. | [anchor 1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [anchor 2](https://huggingface.co/black-forest-labs/FLUX.2-dev), [anchor 3](https://github.com/black-forest-labs/flux2/blob/main/src/flux2/model.py) |
+| `marengo-embed-2.7` | 4 / 4 | Open multimodal-embedding analogues: Qwen3-VL-Embedding 2 B and 8 B. Geometric midpoint sqrt(2 * 8) = 4 B; 4x scenario spread. These omit a separately identifiable audio tower. Dense is an assumption. Context is the text-input limit, not video capacity. | [anchor 1](https://huggingface.co/Qwen/Qwen3-VL-Embedding-2B), [anchor 2](https://huggingface.co/Qwen/Qwen3-VL-Embedding-8B) |
+| `marengo-embed-3.0` | 4 / 4 | Open multimodal-embedding analogues: Qwen3-VL-Embedding 2 B and 8 B. Geometric midpoint sqrt(2 * 8) = 4 B; 4x scenario spread. These omit a separately identifiable audio tower. Dense is an assumption. Context is the text-input limit, not video capacity. | [anchor 1](https://huggingface.co/Qwen/Qwen3-VL-Embedding-2B), [anchor 2](https://huggingface.co/Qwen/Qwen3-VL-Embedding-8B) |
+| `ministral-3b` | 3 / 3 | Published nominal total: 3 B for the original text-only Ministral. Dense and active = total are assumptions; the newer open Ministral 3 3B is a distinct 3.4 B language + 0.4 B vision model, not confirmation of this older architecture. | [Mistral release](https://mistral.ai/news/ministraux/) |
+| `mistral-small-2402` | 22 / 22 | Backward family proxy: 22 B from the later Small-2409 checkpoint; compare Small-3 at 24 B (1.09x). Retain the nearer 22 B generation, not a claim of unchanged weights. | [anchor 1](https://huggingface.co/mistralai/Mistral-Small-Instruct-2409), [anchor 2](https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501) |
+| `multimodalembedding-001` | 0.4 / 0.4 | Use the nominal SigLIP SO400M visual-backbone scale (0.4 B) as a lower-complexity analogue; Qwen3-VL-Embedding-2B supplies a 2 B alternative (5x spread). Select 0.4 B for this older short-text API; text/video towers may add weights. Context is the documented 32-token text limit, not video duration. | [anchor 1](https://huggingface.co/google/siglip-so400m-patch14-384), [anchor 2](https://huggingface.co/Qwen/Qwen3-VL-Embedding-2B) |
+| `palmyra-x4` | 73 / 73 | Publisher-family anchors: X-004 dummy-weight metadata and X-4.3-73B both indicate the 73 B scale, alongside the nominal Fin-70B family. Select 73 B dense as a family proxy, not a disclosed count for this API version. Dummy weights do not establish open-weight status. The 70-73 B comparison is correlated family evidence; X5 may differ substantially. | [anchor 1](https://huggingface.co/Writer/palmyra-x-004-dummy-weights), [anchor 2](https://huggingface.co/Writer/Palmyra-X-4.3-73B), [anchor 3](https://huggingface.co/Writer/Palmyra-Fin-70B-32K) |
+| `palmyra-x5` | 73 / 73 | Publisher-family anchors: X-004 dummy-weight metadata and X-4.3-73B both indicate the 73 B scale, alongside the nominal Fin-70B family. Select 73 B dense as a family proxy, not a disclosed count for this API version. Dummy weights do not establish open-weight status. The 70-73 B comparison is correlated family evidence; X5 may differ substantially. | [anchor 1](https://huggingface.co/Writer/palmyra-x-004-dummy-weights), [anchor 2](https://huggingface.co/Writer/Palmyra-X-4.3-73B), [anchor 3](https://huggingface.co/Writer/Palmyra-Fin-70B-32K) |
+| `pegasus-1.2` | 32 / 32 | Open video-understanding analogues: Qwen2.5-VL 7 B and 72 B (10.3x spread); select its intermediate 32 B checkpoint rather than extrapolating from price per video second. Context is the published 2000-token text-prompt limit, not a joint video/text window. | [anchor 1](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct), [anchor 2](https://huggingface.co/Qwen/Qwen2.5-VL-32B-Instruct), [anchor 3](https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct) |
+| `sora-2` | 14 / 14 | Open video-generator analogues: Wan2.1 14 B and HunyuanVideo 13 B (1.08x spread); sqrt(14 * 13) = 13.49, conservatively rounded upward to 14 B dense. Core video generator only: separate audio/encoder components are not quantified. Fast does not automatically mean fewer weights. This narrow anchor spread is not a confidence interval for the proprietary model. | [anchor 1](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B), [anchor 2](https://huggingface.co/tencent/HunyuanVideo) |
+| `text-embedding-005` | 1.2 / 1.2 | Google Gecko research model (1.2 B) is the preferred same-vendor text-embedding analogue. E5-large (0.56 B) supplies a lower cross-vendor scenario (2.14x spread); it does not override the nearer-family proxy. No official link from this exact API revision to Gecko weights is asserted. | [anchor 1](https://arxiv.org/abs/2403.20327), [anchor 2](https://huggingface.co/intfloat/multilingual-e5-large) |
+| `text-multilingual-embedding-002` | 1.2 / 1.2 | Google Gecko research model (1.2 B) is the preferred same-vendor text-embedding analogue. E5-large (0.56 B) supplies a lower cross-vendor scenario (2.14x spread); it does not override the nearer-family proxy. No official link from this exact API revision to Gecko weights is asserted. | [anchor 1](https://arxiv.org/abs/2403.20327), [anchor 2](https://huggingface.co/intfloat/multilingual-e5-large) |
+| `tts-1` | 4 / 4 | Open TTS analogue: Voxtral-TTS 4 B; select 4 B dense. No comparable token-price calibration: the API bills characters, not tokens. | [anchor 1](https://huggingface.co/mistralai/Voxtral-4B-TTS-2603), [anchor 2](https://developers.openai.com/api/docs/models/tts-1), [anchor 3](https://developers.openai.com/api/docs/models/tts-1-hd) |
+| `tts-1-hd` | 6 / 6 | Two scenarios: unchanged TTS-1-sized weights at 4 B, or price scaling 4 * ($30 / $15) = 8 B using dollars per million characters, not tokens. sqrt(4 * 8) = 5.66, rounded to 6 B dense. A 2x spread is a sensitivity range, not measured uncertainty. | [anchor 1](https://huggingface.co/mistralai/Voxtral-4B-TTS-2603), [anchor 2](https://developers.openai.com/api/docs/models/tts-1), [anchor 3](https://developers.openai.com/api/docs/models/tts-1-hd) |
+| `veo-3.0` | 14 / 14 | Open video-generator analogues: Wan2.1 14 B and HunyuanVideo 13 B (1.08x spread); sqrt(14 * 13) = 13.49, conservatively rounded upward to 14 B dense. Core video generator only: separate audio/encoder components are not quantified. Fast does not automatically mean fewer weights. This narrow anchor spread is not a confidence interval for the proprietary model. | [anchor 1](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B), [anchor 2](https://huggingface.co/tencent/HunyuanVideo) |
+| `veo-3.0-fast` | 14 / 14 | Open video-generator analogues: Wan2.1 14 B and HunyuanVideo 13 B (1.08x spread); sqrt(14 * 13) = 13.49, conservatively rounded upward to 14 B dense. Core video generator only: separate audio/encoder components are not quantified. Fast does not automatically mean fewer weights. This narrow anchor spread is not a confidence interval for the proprietary model. | [anchor 1](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B), [anchor 2](https://huggingface.co/tencent/HunyuanVideo) |
+
+Public capability/context sources are also linked in each `models.json` entry. For image proxies, `hidden_dimension` describes an assumed core-generator width, not an embedding/output image dimension. No image price-per-pixel or video price-per-second was treated as a text-token price.
+
+## Unresolved Historical Estimates
+
+94 existing entries still lack a version-specific numerical source verified in this pass. Their values were not changed or relabelled as independent estimates. Official API/catalog URLs establish public capabilities, not hidden weights. This audit URL documents uncertainty; it is not numerical evidence.
+
+Generic LifeArchitect/ApXML references were removed from these entries unless retained for a historical corpus estimate. A remaining LifeArchitect URL in this section is therefore not an attribution for the parameter values below. Corpus values in these entries were not revalidated in this pass.
+
+| Model | Retained active | Retained total |
+| --- | ---: | ---: |
+| `ada` | 0.35 | 0.35 |
+| `amazon-nova-2-lite` | 64 | 64 |
+| `amazon-nova-2-sonic` | 24 | 24 |
+| `amazon-nova-canvas` | 12 | 12 |
+| `amazon-nova-lite` | 32 | 32 |
+| `amazon-nova-micro` | 8 | 8 |
+| `amazon-nova-multimodal-embeddings` | 8 | 8 |
+| `amazon-nova-reel` | 24 | 24 |
+| `amazon-nova-sonic` | 20 | 20 |
+| `amazon-rerank-1.0` | 1 | 1 |
+| `amazon-titan-image-generator-g1-v2` | 5 | 5 |
+| `amazon-titan-multimodal-embeddings-g1` | 2 | 2 |
+| `amazon-titan-text-embeddings-v1` | 1.5 | 1.5 |
+| `amazon-titan-text-embeddings-v2` | 1.5 | 1.5 |
+| `amazon-titan-text-large` | 13 | 13 |
+| `babbage` | 1.3 | 1.3 |
+| `babbage-002` | 1.3 | 1.3 |
+| `claude` | 175 | 175 |
+| `claude-3-haiku` | 20 | 20 |
+| `claude-3-sonnet` | 70 | 70 |
+| `claude-instant` | 13 | 13 |
+| `code-davinci-002` | 175 | 175 |
+| `cohere-embed-english-v3` | 0.4 | 0.4 |
+| `cohere-embed-multilingual-3` | 1 | 1 |
+| `cohere-embed-multilingual-v3` | 0.8 | 0.8 |
+| `cohere-embed-v4` | 7 | 7 |
+| `cohere-rerank-3.5` | 0.6 | 0.6 |
+| `cohere-rerank-v4.0-fast` | 0.6 | 0.6 |
+| `cohere-rerank-v4.0-pro` | 7 | 7 |
+| `curie` | 6.7 | 6.7 |
+| `davinci` | 175 | 175 |
+| `davinci-002` | 175 | 175 |
+| `embedding-ada` | 0.35 | 0.35 |
+| `gemini-3.5-transcribe` | 8 | 8 |
+| `gemini-embedding-2` | 0.7 | 0.7 |
+| `gpt-4o-realtime` | 55 | 220 |
+| `gpt-4o-realtime-audio` | 55 | 220 |
+| `gpt-5.1-chat` | 400 | 2000 |
+| `gpt-5.1-codex` | 400 | 2000 |
+| `gpt-5.1-codex-max` | 480 | 1920 |
+| `gpt-5.1-codex-mini` | 40 | 160 |
+| `gpt-5.2-chat` | 600 | 3000 |
+| `gpt-5.2-codex` | 600 | 3000 |
+| `gpt-5.3-codex` | 600 | 3000 |
+| `gpt-5.5-pro` | 1200 | 6000 |
+| `gpt-5.6-luna` | 60 | 300 |
+| `gpt-5.6-terra` | 300 | 1500 |
+| `gpt-audio` | 40 | 160 |
+| `gpt-audio-1.5` | 80 | 320 |
+| `gpt-audio-mini` | 12 | 48 |
+| `gpt-image-1` | 40 | 160 |
+| `gpt-image-1-mini` | 12 | 48 |
+| `gpt-image-1.5` | 64 | 256 |
+| `gpt-image-2` | 80 | 320 |
+| `gpt-live-transcribe` | 8 | 8 |
+| `gpt-realtime` | 40 | 160 |
+| `gpt-realtime-1.5` | 80 | 320 |
+| `gpt-realtime-2` | 80 | 320 |
+| `gpt-realtime-2.1` | 80 | 320 |
+| `gpt-realtime-2.1-mini` | 24 | 96 |
+| `gpt-realtime-mini` | 12 | 48 |
+| `gpt-realtime-translate` | 8 | 8 |
+| `gpt-realtime-whisper` | 8 | 8 |
+| `gpt-transcribe` | 8 | 8 |
+| `imagen-3` | 10 | 10 |
+| `imagen-4` | 16 | 16 |
+| `mistral large` | 123 | 123 |
+| `mistral-document-ai-2505` | 24 | 24 |
+| `mistral-document-ai-2512` | 24 | 24 |
+| `mistral-ocr-2505` | 24 | 24 |
+| `mistral-ocr-4` | 24 | 24 |
+| `nova-lite` | 40 | 40 |
+| `nova-micro` | 1 | 1 |
+| `nvr-tts-de-de` | 1.2 | 1.2 |
+| `nvr-tts-en-us` | 1.2 | 1.2 |
+| `nvr-tts-es-es` | 1.2 | 1.2 |
+| `nvr-tts-it-it` | 1.2 | 1.2 |
+| `o3-pro` | 120 | 600 |
+| `text-ada-001` | 0.35 | 0.35 |
+| `text-babbage-001` | 1.3 | 1.3 |
+| `text-curie-001` | 6.7 | 6.7 |
+| `text-davinci-002` | 175 | 175 |
+| `text-embedding-3` | 6.7 | 6.7 |
+| `text-embedding-3-large` | 1.5 | 1.5 |
+| `text-embedding-3-small` | 0.35 | 0.35 |
+| `text-embedding-ada-002` | 0.35 | 0.35 |
+| `titan-embeddings-g1` | 1 | 1 |
+| `titan-embeddings-v2` | 1 | 1 |
+| `titan-text-g1-express` | 15 | 15 |
+| `titan-text-g1-lite` | 5 | 5 |
+| `veo-3.1` | 32 | 32 |
+| `veo-3.1-fast` | 24 | 24 |
+| `veo-3.1-lite` | 16 | 16 |
+| `voxtral-mini-transcribe` | 4 | 4 |
+
+Known legacy duplicate conflicts were not silently resolved: `nova-lite` / `amazon-nova-lite`, `nova-micro` / `amazon-nova-micro`, and the two Cohere multilingual-v3 entries still have conflicting historical estimates. Older names are retained for compatibility; do not count these as distinct underlying models. Canonical consolidation requires a separate compatibility decision.

package/data/country/country-to-country-distances.csv

@@ -4015,7 +4015,7 @@ BA,AU,13970.39
 BA,AW,8859.408
 BA,AX,1807.98
 BA,AZ,2490.744
-BA,BA,NaN
+BA,BA,0
 BA,BB,7976.462
 BA,BD,6853.78
 BA,BE,1234.853
@@ -9788,7 +9788,7 @@ CD,BY,6101.079
 CD,BZ,11535.575
 CD,CA,11949.183
 CD,CC,9028.744
-CD,CD,NaN
+CD,CD,0
 CD,CF,948.694
 CD,CG,59.144
 CD,CH,5283.43
@@ -21334,7 +21334,7 @@ GN,GH,978.227
 GN,GI,2945.357
 GN,GL,7221.193
 GN,GM,724.376
-GN,GN,NaN
+GN,GN,0
 GN,GP,5701.455
 GN,GQ,2391.505
 GN,GR,4500.487
@@ -24346,7 +24346,7 @@ HR,GY,8382.283
 HR,HK,8926.871
 HR,HM,12243.07
 HR,HN,9685.091
-HR,HR,NaN
+HR,HR,0
 HR,HT,8334.696
 HR,HU,403.092
 HR,ID,10753.235
@@ -25099,7 +25099,7 @@ ID,HN,17282.21
 ID,HR,10753.235
 ID,HT,17883.086
 ID,HU,10405.813
-ID,ID,NaN
+ID,ID,0
 ID,IE,12138.947
 ID,IL,9013.892
 ID,IM,11880.4
@@ -27358,7 +27358,7 @@ IT,IO,8039.636
 IT,IQ,2887.673
 IT,IR,3753.56
 IT,IS,3243.384
-IT,IT,NaN
+IT,IT,0
 IT,JE,1402.34
 IT,JM,8665.192
 IT,JO,2454.056
@@ -32629,7 +32629,7 @@ LR,LB,5558.67
 LR,LC,5692.202
 LR,LI,4874.098
 LR,LK,9931.872
-LR,LR,NaN
+LR,LR,0
 LR,LS,5676.881
 LR,LT,6172.828
 LR,LU,5038.599

package/data/country/country-to-country-distances.json

@@ -4015,7 +4015,7 @@
   "BAAW": 8859.408,
   "BAAX": 1807.98,
   "BAAZ": 2490.744,
-  "BABA": null,
+  "BABA": 0,
   "BABB": 7976.462,
   "BABD": 6853.78,
   "BABE": 1234.853,
@@ -9788,7 +9788,7 @@
   "CDBZ": 11535.575,
   "CDCA": 11949.183,
   "CDCC": 9028.744,
-  "CDCD": null,
+  "CDCD": 0,
   "CDCF": 948.694,
   "CDCG": 59.144,
   "CDCH": 5283.43,
@@ -21334,7 +21334,7 @@
   "GNGI": 2945.357,
   "GNGL": 7221.193,
   "GNGM": 724.376,
-  "GNGN": null,
+  "GNGN": 0,
   "GNGP": 5701.455,
   "GNGQ": 2391.505,
   "GNGR": 4500.487,
@@ -24346,7 +24346,7 @@
   "HRHK": 8926.871,
   "HRHM": 12243.07,
   "HRHN": 9685.091,
-  "HRHR": null,
+  "HRHR": 0,
   "HRHT": 8334.696,
   "HRHU": 403.092,
   "HRID": 10753.235,
@@ -25099,7 +25099,7 @@
   "IDHR": 10753.235,
   "IDHT": 17883.086,
   "IDHU": 10405.813,
-  "IDID": null,
+  "IDID": 0,
   "IDIE": 12138.947,
   "IDIL": 9013.892,
   "IDIM": 11880.4,
@@ -27358,7 +27358,7 @@
   "ITIQ": 2887.673,
   "ITIR": 3753.56,
   "ITIS": 3243.384,
-  "ITIT": null,
+  "ITIT": 0,
   "ITJE": 1402.34,
   "ITJM": 8665.192,
   "ITJO": 2454.056,
@@ -32629,7 +32629,7 @@
   "LRLC": 5692.202,
   "LRLI": 4874.098,
   "LRLK": 9931.872,
-  "LRLR": null,
+  "LRLR": 0,
   "LRLS": 5676.881,
   "LRLT": 6172.828,
   "LRLU": 5038.599,

package/data/country/region-to-region-distances.csv

@@ -5039,7 +5039,7 @@ BA,AU,13970.39
 BA,AW,8859.408
 BA,AX,1807.98
 BA,AZ,2490.744
-BA,BA,NaN
+BA,BA,0
 BA,BB,7976.462
 BA,BD,6853.78
 BA,BE,1234.853
@@ -12914,7 +12914,7 @@ CA-NB,CA,2904.225
 CA-NB,CA-AB,3447.996
 CA-NB,CA-BC,4079.05
 CA-NB,CA-MB,2305.205
-CA-NB,CA-NB,NaN
+CA-NB,CA-NB,0
 CA-NB,CA-NL,903.459
 CA-NB,CA-NS,313.208
 CA-NB,CA-NT,3659.678
@@ -16379,7 +16379,7 @@ CD,CA-QC,9868.765
 CD,CA-SK,11968.269
 CD,CA-YT,12508.011
 CD,CC,9028.744
-CD,CD,NaN
+CD,CD,0
 CD,CF,948.694
 CD,CG,59.144
 CD,CH,5283.43
@@ -30869,7 +30869,7 @@ GN,GH,978.227
 GN,GI,2945.357
 GN,GL,7221.193
 GN,GM,724.376
-GN,GN,NaN
+GN,GN,0
 GN,GP,5701.455
 GN,GQ,2391.505
 GN,GR,4500.487
@@ -34649,7 +34649,7 @@ HR,GY,8382.283
 HR,HK,8926.871
 HR,HM,12243.07
 HR,HN,9685.091
-HR,HR,NaN
+HR,HR,0
 HR,HT,8334.696
 HR,HU,403.092
 HR,ID,10753.235
@@ -35594,7 +35594,7 @@ ID,HN,17282.21
 ID,HR,10753.235
 ID,HT,17883.086
 ID,HU,10405.813
-ID,ID,NaN
+ID,ID,0
 ID,IE,12138.947
 ID,IL,9013.892
 ID,IM,11880.4
@@ -38429,7 +38429,7 @@ IT,IO,8039.636
 IT,IQ,2887.673
 IT,IR,3753.56
 IT,IS,3243.384
-IT,IT,NaN
+IT,IT,0
 IT,JE,1402.34
 IT,JM,8665.192
 IT,JO,2454.056
@@ -45044,7 +45044,7 @@ LR,LB,5558.67
 LR,LC,5692.202
 LR,LI,4874.098
 LR,LK,9931.872
-LR,LR,NaN
+LR,LR,0
 LR,LS,5676.881
 LR,LT,6172.828
 LR,LU,5038.599
@@ -78119,7 +78119,7 @@ US-AZ,UM,8131.482
 US-AZ,US,1429.883
 US-AZ,US-AK,4415.984
 US-AZ,US-AL,2254.252
-US-AZ,US-AZ,NaN
+US-AZ,US-AZ,0
 US-AZ,US-AR,1764.369
 US-AZ,US-CA,812.748
 US-AZ,US-CO,772.869
@@ -82214,7 +82214,7 @@ US-IA,US-HI,6300.399
 US-IA,US-ID,1772.474
 US-IA,US-IL,338.756
 US-IA,US-IN,610.329
-US-IA,US-IA,NaN
+US-IA,US-IA,0
 US-IA,US-KS,556.004
 US-IA,US-KY,876.495
 US-IA,US-LA,1227.848

package/data/country/region-to-region-distances.json

@@ -5039,7 +5039,7 @@
   "BAAW": 8859.408,
   "BAAX": 1807.98,
   "BAAZ": 2490.744,
-  "BABA": null,
+  "BABA": 0,
   "BABB": 7976.462,
   "BABD": 6853.78,
   "BABE": 1234.853,
@@ -12914,7 +12914,7 @@
   "CA-NBCA-AB": 3447.996,
   "CA-NBCA-BC": 4079.05,
   "CA-NBCA-MB": 2305.205,
-  "CA-NBCA-NB": null,
+  "CA-NBCA-NB": 0,
   "CA-NBCA-NL": 903.459,
   "CA-NBCA-NS": 313.208,
   "CA-NBCA-NT": 3659.678,
@@ -16379,7 +16379,7 @@
   "CDCA-SK": 11968.269,
   "CDCA-YT": 12508.011,
   "CDCC": 9028.744,
-  "CDCD": null,
+  "CDCD": 0,
   "CDCF": 948.694,
   "CDCG": 59.144,
   "CDCH": 5283.43,
@@ -30869,7 +30869,7 @@
   "GNGI": 2945.357,
   "GNGL": 7221.193,
   "GNGM": 724.376,
-  "GNGN": null,
+  "GNGN": 0,
   "GNGP": 5701.455,
   "GNGQ": 2391.505,
   "GNGR": 4500.487,
@@ -34649,7 +34649,7 @@
   "HRHK": 8926.871,
   "HRHM": 12243.07,
   "HRHN": 9685.091,
-  "HRHR": null,
+  "HRHR": 0,
   "HRHT": 8334.696,
   "HRHU": 403.092,
   "HRID": 10753.235,
@@ -35594,7 +35594,7 @@
   "IDHR": 10753.235,
   "IDHT": 17883.086,
   "IDHU": 10405.813,
-  "IDID": null,
+  "IDID": 0,
   "IDIE": 12138.947,
   "IDIL": 9013.892,
   "IDIM": 11880.4,
@@ -38429,7 +38429,7 @@
   "ITIQ": 2887.673,
   "ITIR": 3753.56,
   "ITIS": 3243.384,
-  "ITIT": null,
+  "ITIT": 0,
   "ITJE": 1402.34,
   "ITJM": 8665.192,
   "ITJO": 2454.056,
@@ -45044,7 +45044,7 @@
   "LRLC": 5692.202,
   "LRLI": 4874.098,
   "LRLK": 9931.872,
-  "LRLR": null,
+  "LRLR": 0,
   "LRLS": 5676.881,
   "LRLT": 6172.828,
   "LRLU": 5038.599,
@@ -78119,7 +78119,7 @@
   "US-AZUS": 1429.883,
   "US-AZUS-AK": 4415.984,
   "US-AZUS-AL": 2254.252,
-  "US-AZUS-AZ": null,
+  "US-AZUS-AZ": 0,
   "US-AZUS-AR": 1764.369,
   "US-AZUS-CA": 812.748,
   "US-AZUS-CO": 772.869,
@@ -82214,7 +82214,7 @@
   "US-IAUS-ID": 1772.474,
   "US-IAUS-IL": 338.756,
   "US-IAUS-IN": 610.329,
-  "US-IAUS-IA": null,
+  "US-IAUS-IA": 0,
   "US-IAUS-KS": 556.004,
   "US-IAUS-KY": 876.495,
   "US-IAUS-LA": 1227.848,