Model

Inflection-2

CLOSED

Inflection AI · Inflection family · released Nov 22, 2023

Trained on 5,000 H100 GPUs at ~10^25 FLOPs; billed at launch as the second-most-capable LLM in the world, beating PaLM 2 Large.

ReasoningCodingVisionFunction callingTool useAgentic

646.1

Elo · rank #275

Parameters

Undisclosed

Active params

—

Context

8K tokens

Architecture

Dense Transformer (architecture undisclosed; trained at ~10^25 FLOPs)

License

Proprietary

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

ARC-ChallengeReasoning70.7%#47

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning91.3%#5

best: Phi-3-medium (14B) · 97.7%

GSM8KMath81.4%#76

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning89.0%#12

best: Claude 3 Opus · 95.4%

HumanEvalCoding44.5%#132

best: Claude Opus 4.5 · 99.4%

MBPPCoding53.0%#67

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLUKnowledge79.6%#93

best: OpenAI o3 · 92.9%

PIQAReasoning84.2%#12

best: GPT-4o mini · 93.1%

TriviaQAKnowledge87.2%#6

best: Sarvam-1 (2B) · 90.6%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

—

Output / M tok

—

Inflection family

Elo progression across releases

Inflection-1Jun 2023513.0 Inflection-2Nov 2023646.1 Inflection-2.5Mar 2024870.8

Resources

DOCSinflection.ai/inflection-2

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