Model
Model explorer

CodeT5+ 16B

OPEN
Salesforce · CodeT5+ family · released May 13, 2023

Encoder-decoder code LLM family (220M-16B); InstructCodeT5+ 16B tuned variant hit open-source SOTA HumanEval pass@1 in 2023 (non-commercial license)

ReasoningCodingVisionFunction callingTool useAgentic
41.1
Elo · unrated
Parameters
16B
Active params
16B (dense)
Context
2K tokens
Architecture
Encoder-decoder (shallow encoder / deep decoder), initialized from CodeGen-350M-mono encoder + CodeGen-16B-mono decoder
License
BSD-3-Clause
Languages
—
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
HumanEvalCoding30.9%#152
best: Claude Opus 4.5 · 99.4%
Run it locally
VRAM @ Q4
—
VRAM @ FP16
32 GB
Fits on (Q4)
Multi-node cluster required
Throughput data unavailable.
Quantizations
—
Fine-tune it
Permissive
QLoRA12.1 GB1× RTX 4070 Ti 16GB
LoRA35.3 GB1× A100 80GB
Full fine-tune258.0 GB2× H200 141GB
QLoRA SFT on ~10k samples ≈ $5.85 (1× RTX 4070 Ti 16GB)
Plan a fine-tune →
CodeT5+ family
Elo progression across releases
CodeT5+ 16BMay 202341.1
Resources
HFSalesforce/codet5p-16bGHsalesforce/CodeT5
API price weights · each benchmark row carries its own source badge (see methodology)