Skip to content
fedi.software

DeepSeek V3.2 Exp in fedi-index

The score in every slice, what it is made of, value, local run and history.

fedi-index · General

B 1,395 ±8

Percentile 65 · #102

People prefer / Solves tasks

1,406 / 1,383

Human votes vs benchmarks, Elo-eq

Value

+46

$0.305 per 1M tokens, median of 7 providers

On your hardware

Does not fit the reference setups.

Score decomposition

B General 1,395 [1,387–1,403] Percentile 65 · #102 · components: 7
Source Component Value Elo-eq Weight Measured Variant
Epoch AI eci 145 ±1.17 1,398 ±9 41.8% 2025-09-29 —
LMArena text/overall 1,425 ±3.35 1,425 ±3 15.7% 2026-10-02 thinking
LMArena text/hard_prompts 1,429 ±4.03 1,421 ±4 15.6% 2026-10-02 —
LMArena text/instruction_following 1,401 ±5.22 1,414 ±5 7.4% 2026-10-02 —
LMArena text/expert 1,420 ±14 1,411 ±12 5.1% 2026-10-02 thinking
LMArena text/longer_query 1,419 ±5.43 1,418 ±5 3.7% 2026-10-02 —
LMArena text/multi_turn 1,422 ±6.60 1,419 ±6 3.6% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Coding 1,379 [1,374–1,385] Percentile 57 · #121 · components: 2 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena webdev/overall 1,272 ±6.87 1,385 ±2 47.9% 2026-10-01 —
LMArena text/coding 1,439 ±6.79 1,416 ±6 41.9% 2026-10-02 thinking

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Math 1,381 [1,364–1,397] Percentile 57 · #110 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/math 1,420 ±10 1,412 ±10 84.9% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Writing 1,412 [1,402–1,421] Percentile 70 · #82 · components: 2 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/creative_writing 1,404 ±7.40 1,433 ±8 59.5% 2026-10-02 —
LMArena text/longer_query 1,419 ±5.43 1,418 ±5 33.5% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Language: 繁體中文 1,404 [1,385–1,423] Percentile 60 · #91 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/chinese 1,462 ±13 1,423 ±11 91.4% 2026-10-02 thinking

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Language: English 1,416 [1,407–1,426] Percentile 71 · #81 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/english 1,441 ±4.85 1,431 ±5 93.5% 2026-10-02 thinking

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Language: Español 1,410 [1,385–1,436] Percentile 53 · #70 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/spanish 1,440 ±16 1,436 ±15 89.1% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Language: Polski 1,415 [1,392–1,437] Percentile 56 · #58 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/polish 1,435 ±15 1,437 ±12 90.4% 2026-10-02 thinking

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Language: Русский 1,398 [1,380–1,416] Percentile 60 · #95 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/russian 1,410 ±11 1,416 ±10 91.7% 2026-10-02 thinking

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

History

fedi-index by month
The same in a table
Model 2026-10
DeepSeek V3.2 Exp 1,395 B

Badge

Show the model's fedi-index on your site or in a README. The badge updates itself; it links to this page.

fedi-index: DeepSeek V3.2 Exp

Data CC BY 4.0: keep the link to fedi.software.