rewirebio.iobenchmarks

Biological models

Understand model architectures, biological inputs, access requirements and the evidence from linked evaluations.

374 model records in release 2026-11-21-c8bea25e5abb (release manifest, gzip). Showing 49–72; page 3 of 16.

Evaluated configurations, methods and pipelines are listed separately. Names alone do not establish equivalent models or checkpoints.

  • ChemLLM-7B

    Protein function

    ChemLLM-7B chemistry large language model, prompted for protein property questions.

  • Chroma

    Protein structure

    Diffusion-based protein backbone generation method evaluated in Scaffold-Lab (released 2023-11).

  • ChromBPNet

    DNA and genomes

    ChromBPNet predicts base-resolution chromatin accessibility while modeling assay-specific enzyme bias separately.

  • CI-SpliceAI

    DNA and genomes

    SpliceAI reimplementation retrained on a collapsed isoform set of manually curated GENCODE HAVANA splice sites from all chromosomes.

  • Claude 3 Haiku

    Cells and tissues

    Large multimodal model with vision capabilities, the fastest and least expensive of Anthropic's Claude 3 family.

  • Claude 3 Opus

    Cells and tissues

    Large multimodal model with vision capabilities, the most capable of Anthropic's Claude 3 family.

  • Claude 3 Sonnet

    Cells and tissues

    Large multimodal model with vision capabilities in Anthropic's Claude 3 family, positioned between Opus and Haiku in skills and speed.

  • Claude 3.5 Sonnet

    Cells and tissues

    Large multimodal model from Anthropic, an evolution of the Claude 3 family with improved coding and visual processing.

  • Claude Haiku 4.5

    Cells and tissues

    Anthropic's hosted large language model Claude Haiku 4.5.

  • Claude Mythos 5.1

    DNA and genomes

    Large language model served as claude-mythos-5-1 over the Anthropic API, which plans molecular cloning experiments when driven by an agent harness and which also scores other systems' construction plans as a second judge.

  • Claude Opus 4.8

    Cells and tissues

    Anthropic's hosted large language model Claude Opus 4.8.

  • Claude Opus 5.5

    DNA and genomes

    Large language model served as claude-opus-5-5 over the Anthropic API, which plans molecular cloning experiments when driven by an agent harness.

  • Claude Sonnet 4

    Molecular interactions · Protein function · Single-cell omics

    Anthropic language model, used as the LLM backend for Biomni in BioML-bench.

  • Claude Sonnet 4.6

    Cells and tissues

    Anthropic's hosted large language model Claude Sonnet 4.6.

  • Claude v1

    Cells and tissues

    Anthropic's original Claude AI assistant model, offered through a chat interface and an API for conversational and text-processing tasks.

  • CMLM

    Proteins and complexes · Protein structure

    CMLM designs a sequence for a supplied backbone by conditional masked language modelling, decoding all positions in parallel from one categorical distribution.

  • CNN

    Protein function

    Convolutional neural network over the one-hot encoded sequence.

  • CNN-RNN

    Protein function

    Convolutional network followed by a recurrent layer over the sequence.

  • CodonBERT

    DNA and genomes

    BERT-style language model of mRNA coding sequences at codon resolution.

  • CPA

    Cells and tissues · Single-cell omics

    Perturbation-response model the authors benchmarked against (main text paragraph 2); the authors say it was not designed to predict the effects of unseen perturbations (main text paragraphs 16 and 20).

  • DeCodon

    DNA and genomes

    Decoder-only codon language model.

  • Deep Prot-T5

    Protein function

    DeepProt-T5: Prot-T5-XL fine-tuned per task, introduced by the DeepProtein authors.

  • DeepAffinity

    Molecular interactions

    DeepAffinity predicts compound-protein affinity from a compound SMILES string and a structure property-annotated protein sequence (SPS), using pretrained recurrent seq2seq encoders followed by convolutional layers.

  • DeepDTA

    Molecular interactions

    DeepDTA predicts drug-target binding affinity from a ligand SMILES string and a protein sequence, using a separate 1D convolutional encoder for each.

This index reflects a dated catalogue, not an exhaustive census. Source checking does not mean independent reproduction; compare results only under compatible protocols, datasets and metrics.