Llama 3.1 405B Instruct (BF16) in fedi-index
The score in every slice, what it is made of, value, local run and history.
fedi-index · General
C 1,268 ±8
Percentile 31 · #205
People prefer / Solves tasks
1,275 / 1,257
Human votes vs benchmarks, Elo-eq
Value
-144
$3.50 per 1M tokens, median of 1 providers
On your hardware
Closed weights: cloud only.
Score decomposition
C General 1,268 [1,261–1,276] Percentile 31 · #205 · components: 8
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| Epoch AI |
eci
|
128 ±1.47 | 1,263 ±12 | 32.6% | 2024-07-23 | — |
| LMArena |
text/overall
|
1,284 ±1.89 | 1,284 ±2 | 17.1% | 2026-11-11 | bf16 |
| LMArena |
text/hard_prompts
|
1,269 ±3.03 | 1,277 ±3 | 16.8% | 2026-11-11 | bf16 |
| LMArena |
text/instruction_following
|
1,258 ±2.68 | 1,279 ±2 | 8.4% | 2026-11-11 | bf16 |
| LMArena |
text/expert
|
1,243 ±5.81 | 1,266 ±5 | 7.9% | 2026-11-11 | — |
| LMArena |
text/multi_turn
|
1,297 ±3.91 | 1,305 ±4 | 4.1% | 2026-11-11 | bf16 |
| LMArena |
text/longer_query
|
1,266 ±4.08 | 1,275 ±4 | 4.1% | 2026-11-11 | bf16 |
| Epoch AI |
gpqa_diamond
decay ×0.25 |
0.51 ±0.03 | 1,277 ±12 | 1.3% | 2025-01-27 | — |
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
C Coding 1,275 [1,269–1,281] Percentile 29 · #203 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/coding
|
1,291 ±3.81 | 1,284 ±3 | 88.4% | 2026-11-11 | bf16 |
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
C Math 1,270 [1,264–1,277] Percentile 32 · #173 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/math
|
1,281 ±3.86 | 1,279 ±4 | 88.3% | 2026-11-11 | — |
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
C Writing 1,278 [1,273–1,284] Percentile 35 · #173 · components: 2 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/creative_writing
|
1,262 ±3.92 | 1,287 ±4 | 62.4% | 2026-11-11 | — |
| LMArena |
text/longer_query
|
1,266 ±4.08 | 1,275 ±4 | 31.4% | 2026-11-11 | bf16 |
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
D Language: 繁體中文 1,246 [1,237–1,254] Percentile 23 · #168 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/chinese
|
1,242 ±5.64 | 1,248 ±4 | 93.6% | 2026-11-11 | bf16 |
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
D Language: Deutsch 1,269 [1,254–1,285] Percentile 19 · #118 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/german
|
1,252 ±9.94 | 1,275 ±9 | 92.3% | 2026-11-11 | bf16 |
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
C Language: English 1,289 [1,285–1,294] Percentile 36 · #172 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/english
|
1,312 ±2.35 | 1,295 ±2 | 94.0% | 2026-11-11 | bf16 |
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
D Language: Español 1,262 [1,237–1,287] Percentile 9 · #136 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/spanish
|
1,260 ±15 | 1,269 ±14 | 89.3% | 2026-11-11 | bf16 |
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
D Language: Français 1,268 [1,242–1,293] Percentile 11 · #126 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/french
|
1,279 ±16 | 1,275 ±14 | 89.3% | 2026-11-11 | bf16 |
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
D Language: 日本語 1,277 [1,263–1,292] Percentile 16 · #107 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/japanese
|
1,208 ±11 | 1,283 ±8 | 92.6% | 2026-11-11 | bf16 |
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
D Language: 한국어 1,266 [1,250–1,282] Percentile 12 · #121 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/korean
|
1,184 ±11 | 1,271 ±9 | 92.3% | 2026-11-11 | — |
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
C Language: Русский 1,280 [1,274–1,287] Percentile 27 · #169 · components: 1 provisional
| Source | Component | Value | Elo-eq | Weight | Measured | Variant |
|---|---|---|---|---|---|---|
| LMArena |
text/russian
|
1,265 ±4.04 | 1,286 ±4 | 93.8% | 2026-11-11 | — |
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
The same in a table
| Model | 2026-11 |
|---|---|
| Llama 3.1 405B Instruct (BF16) | 1,268 C |
Badge
Show the model's fedi-index on your site or in a README. The badge updates itself; it links to this page.
Data CC BY 4.0: keep the link to fedi.software.