Model

Inflection-2.5

CLOSED

Inflection AI · Inflection family · released Mar 7, 2024

Reached ~94% of GPT-4's average IQ-benchmark performance using only 40% of GPT-4's training compute; Inflection's last flagship before the March 2024 Microsoft talent exodus.

ReasoningCodingVisionFunction callingTool useAgentic

870.8

Elo · rank #243

Parameters

Undisclosed

Active params

—

Context

8K tokens

Architecture

Dense Transformer (architecture undisclosed)

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-ChallengeReasoning94.6%#10

best: Llama 3.1 405B · 96.9%

BIG-Bench HardReasoning82.2%#27

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning38.4%#226

best: GPT-5.6 · 94.6%

GSM8KMath86.3%#59

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning90.2%#7

best: Claude 3 Opus · 95.4%

HumanEvalCoding73.8%#71

best: Claude Opus 4.5 · 99.4%

MATH-500Math43.1%#145

best: GPT-5 · 99.4%

MBPPCoding73.9%#27

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

MMLUKnowledge85.5%#54

best: OpenAI o3 · 92.9%

MT-BenchHuman preference8.41#28

best: Hunyuan-Large (A52B) · 9.4

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-2-5

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