LLM4HEP / COMPLETE_MODEL_VERSIONS.md
ho22joshua's picture
initial commit
cfcbbc8

Complete Model Version Information

Discovered via CBORG API Testing - October 29, 2025

This document shows the complete mapping from CBORG model aliases to their underlying versions, including all version dates discovered through API testing.


Models with Version Dates

Anthropic Claude Models

Model Alias Display Name Underlying Version Version Date
anthropic/claude-haiku:latest Claude Haiku 4.5 (2025-10-01) claude-haiku-4-5@20251001 Oct 1, 2025
anthropic/claude-opus:latest Claude Opus 4.1 (2025-08-05) us.anthropic.claude-opus-4-1-20250805-v1:0 Aug 5, 2025
anthropic/claude-sonnet:latest Claude Sonnet 4.5 (2025-09-29) claude-sonnet-4-5@20250929 Sep 29, 2025
claude-3-5-haiku-latest Claude 3.5 Haiku (2024-10-22) claude-3-5-haiku@20241022 Oct 22, 2024

OpenAI Models (via Azure)

Model Alias Display Name Underlying Version Version Date
openai/gpt-5 GPT-5 (2025-08-07) gpt-5-2025-08-07 Aug 7, 2025
openai/gpt-5-mini GPT-5 Mini (2025-08-07) gpt-5-mini-2025-08-07 Aug 7, 2025
openai/o:latest O3 (2025-04-16) azure/o3-2025-04-16 Apr 16, 2025
openai/o3 O3 (2025-04-16) azure/o3-2025-04-16 Apr 16, 2025
openai/o3-mini O3 Mini (2025-01-31) azure/o3-mini-2025-01-31 Jan 31, 2025
openai/o4-mini O4 Mini (2025-04-16) azure/o4-mini-2025-04-16 Apr 16, 2025

Key Finding: Both openai/o:latest and openai/o3 map to the same model version (2025-04-16)


Models with Model Size Information

AWS Llama Models

Model Alias Display Name Underlying Version
aws/llama-4-maverick Llama-4 Maverick (17B) us.meta.llama4-maverick-17b-instruct-v1:0
aws/llama-4-scout Llama-4 Scout (17B) us.meta.llama4-scout-17b-instruct-v1:0

Key Finding: Both models are 17 billion parameter variants

GCP Models

Model Alias Display Name Underlying Version
gcp/qwen-3 Qwen-3 (235B) qwen/qwen3-235b-a22b-instruct-2507-maas

Key Finding: This is a massive 235 billion parameter model


Google Gemini Models

Model Alias Display Name Underlying Version Notes
google/gemini:latest Gemini 2.5 Pro gemini-2.5-pro Latest generation
google/gemini-flash Gemini 2.5 Flash gemini-2.5-flash Fast variant
gemini-2.0-flash-lite Gemini 2.0 Flash Lite (no alias - direct name) Lightweight variant

xAI Grok Models

Model Alias Display Name Underlying Version Notes
xai/grok:latest Grok-3 grok-3 Latest generation
xai/grok-mini Grok Mini (rate limited during test) Smaller variant
xai/grok-code-fast-1 Grok Code Fast 1 (rate limited during test) Code-focused fast variant

Other Models

Model Alias Display Name Underlying Version Notes
gpt-oss-120b GPT-OSS-120B hosted_vllm/hosted_vllm/gpt-oss-120b Open source, hosted via vLLM
gpt-5-codex GPT-5 Codex (not accessible during test) Code-focused variant
deepseek-r1 DeepSeek-R1 MAI-DS-R1 DeepSeek reasoning model

Key Insights

Version Date Patterns

  1. Most Recent Claude Models: September-October 2025

    • Sonnet 4.5: Sep 29, 2025
    • Haiku 4.5: Oct 1, 2025
    • Opus 4.1: Aug 5, 2025
  2. Most Recent OpenAI Models: April-August 2025

    • GPT-5: Aug 7, 2025
    • O4 Mini: Apr 16, 2025
    • O3: Apr 16, 2025
    • O3 Mini: Jan 31, 2025
  3. Older Models Still in Use:

    • Claude 3.5 Haiku: Oct 22, 2024 (over a year old)

Model Sizes Discovered

  • 235B parameters: Qwen-3 (largest)
  • 120B parameters: GPT-OSS-120B
  • 17B parameters: Llama-4 Maverick, Llama-4 Scout

:latest Aliases

All :latest suffixes have been resolved:

  • anthropic/claude-*:latest → Specific dated versions
  • google/gemini:latest → gemini-2.5-pro
  • xai/grok:latest → grok-3
  • openai/o:latest → azure/o3-2025-04-16

Usage in Notebook

The notebook now displays all these version dates and model sizes in plot titles and legends, making it clear exactly which model versions were used in the experiments.

Example plot titles:

  • "Claude Haiku 4.5 (2025-10-01)" instead of "anthropic/claude-haiku:latest"
  • "O3 (2025-04-16)" instead of "openai/o3"
  • "GPT-5 Mini (2025-08-07)" instead of "openai/gpt-5-mini"
  • "Qwen-3 (235B)" instead of "gcp/qwen-3"

This provides complete transparency about which exact model snapshots were used in your analysis!