Is Llama 3.1 8B Instruct a good deal?

Whether anything in this catalogue beats Llama 3.1 8B Instruct on both quality and price, and what you give up if it does. A computation on the current catalogue, not an opinion.

13 undominated of 136 · Sep 23, 2026

As of Sep 23, 2026, gpt-oss-20b scores higher and costs less than Llama 3.1 8B Instruct for Balanced on LMArena, but it is not a drop-in. You would give up: max output.

Inspect model evidence Compare differences & requirements

Current model

Llama 3.1 8B Instruct

Balanced

3 tokens in per 1 out

gpt-oss-20b scores 100.8 points higher and costs 37% less, but you would give up: max output.

Llama 3.1 8B Instruct

LMArena Elo · higher is better

Scale starts at 1040 Elo

1186.5

Effective $/M · Balanced · lower is better

$0.058/M

gpt-oss-20b

LMArena Elo · higher is better

Scale starts at 1040 Elo

1287.3

Effective $/M · Balanced · lower is better

$0.036/M

Llama 3.1 8B Instruct takes text, returns up to 117,964 tokens from a 131,072-token context, and is listed by 5 sellers.

Compared against 136 rated, priced models on this lens: nothing dominates it outright, 2 more dominate it but give something up. Llama 3.1 8B Instruct scores 1186.5 at $0.06 per million tokens for this mix.

Higher score, lower price, named losses

Not a drop-in. The loss is why these are not a recommendation.

ModelLMArenaEffective $/MYou give up
gpt-oss-20b1287.3$0.036/Mmax output
Mistral Small 31233.5$0.058/Mcontext, max output, tools
Llama 3.1 8B Instruct1186.5 · $0.058/MLMArena EloEffective $/M · Balanced
136 rated, priced standard models. Chartreuse is the frontier. Cobalt is the model you named, when it is not on the staircase.

Open the frontier

The link keeps the model and mix, using the current catalogue. Monthly spend and switching cost stay in this browser tab and are left out of the link.

Saved decision references

Save the model, workload, catalogue date and capability-preservation rule in this browser. Spend, switching cost and bill contents are not saved. No account or notifications.

Constraint: replacements must preserve the model’s capabilities; any losses remain named trade-offs.

Historical decisions cannot be fully replayed from saved references: past prices, scores and capability evidence are not stored.

Evidence & Ask