Model

XGen-7B 8K

OPEN

Salesforce · XGen family · released Jul 3, 2023

General-purpose 7B LLM trained on 1.5T tokens with an 8K context window under Apache-2.0; Salesforce's long-context entry

ReasoningCodingVisionFunction callingTool useAgentic

-290.1

Elo · rank #431

Parameters

Active params

7B (dense)

Context

8K tokens

Architecture

Dense transformer, trained with staged 2K to 4K to 8K sequence-length curriculum, OpenAI Tiktoken tokenizer

License

Apache-2.0

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

ARC-ChallengeReasoning41.2%#145

best: Llama 3.1 405B · 96.9%

HellaSwagReasoning74.2%#115

best: Claude 3 Opus · 95.4%

HumanEvalCoding14.2%#178

best: Claude Opus 4.5 · 99.4%

MMLUKnowledge32.1%#267

best: OpenAI o3 · 92.9%

OpenBookQAReasoning40.2%#43

best: Claude 1 · 90.8%

PIQAReasoning75.5%#70

best: GPT-4o mini · 93.1%

TruthfulQAKnowledge39.1%#84

best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%

WinoGrandeReasoning64.9%#124

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

—

VRAM @ FP16

14 GB

Fits on (Q4)

Multi-node cluster required

Throughput data unavailable.

Quantizations

SqueezeLLM 3-bit

XGen family

Elo progression across releases

XGen-7B 8KJul 2023-290.1

Resources

HFSalesforce/xgen-7b-8k-base GHsalesforce/xgen

API price weights · each benchmark row carries its own source badge (see methodology)