// head_to_head

DeepSeek V4 Pro 0813 vs GPT-5.6 Terra

Published LiveBench scores across all seven categories, live list pricing, context windows, and the measured cost of a point of capability — for both models, side by side.

OpenRouter + LiveBenchAll comparisonsFull leaderboard

Effectively the same quality — DeepSeek V4 Pro 0813 is the cheaper way to get it.

The two are within 0.5 points overall, which is inside the range that effort settings alone move a LiveBench score, so treat them as quality-equivalent. DeepSeek V4 Pro 0813 lists 2.5× cheaper per blended million tokens. When quality ties, cost is the whole decision. DeepSeek V4 Pro 0813 also leads on measured cost per point of capability, at $0.0241 per point.

deepseek

DeepSeek V4 Pro 0813

Blended / 1M
$1.78
Context
1.0M
Released
Aug 12, 2026
Overall score
77.4
reasoningtool callingprompt caching

openai

GPT-5.6 Terra

Blended / 1M
$4.50
Context
1.1M
Released
Jul 9, 2026
Overall score
77.9
reasoningtool callingfile inputimage inputprompt caching

Specs and pricing

MetricDeepSeek V4 Pro 0813GPT-5.6 Terra
LiveBench overall

Mean of the seven LiveBench category scores, 0–100. Higher is better.

77.477.9
Cost per point

Measured benchmark spend divided by overall score — dollars per point of capability.

$0.0241win$0.1939
Blended price / 1M

3:1 input:output mix, the usual shape of production traffic.

$1.78win$4.50
Input price / 1M$1.19win$2.00
Output price / 1M$3.56win$12.00
Cached input / 1M

Price of an input token served from the prompt cache, where the provider publishes one.

$0.040win$0.200
Context window1.0M1.1M
Max output tokens128K

Benchmarks by category

An overall score averages away the thing you probably care about. These are the seven LiveBench categories scored separately — DeepSeek V4 Pro 0813 on top, GPT-5.6 Terra below, both out of 100.

Agentic codingtoo close to call
54.9
54.9
Coding
77.2
78.2
Reasoning
85.8
90.6
Mathematicstoo close to call
95.1
94.9
Data analysistoo close to call
79.2
79.3
Languagetoo close to call
82.1
82.9
Instruction following
67.7
64.6

What each one costs to run

Per-token prices are hard to feel. These are monthly list costs for both models across five workload shapes, using each provider's published cached-input rate where there is one.

WorkloadDeepSeek V4 Pro 0813GPT-5.6 Terra
Support chatbot

1.2K in / 400 out × 200K requests

$487.56/mo$1310.40/mo
RAG assistant

8K in / 600 out × 100K requests

$704.88/mo$1600.00/mo
Coding agent

40K in / 4K out × 20K requests

$592.42/mo$1552.00/mo
Document extraction

20K in / 1.5K out × 50K requests

$1397.88/mo$2810.00/mo
Bulk classification

500 in / 20 out × 5M requests

$2752.20/mo$5300.00/mo
Run these two through the cost calculator

Which should you pick?

You are running this at volume

DeepSeek V4 Pro 0813

Lowest measured cost per point of capability at $0.0241 per point — the gap compounds with every request.

DeepSeek V4 Pro 0813 vs GPT-5.6 Terra FAQ

Which is better, DeepSeek V4 Pro 0813 or GPT-5.6 Terra?

Effectively the same quality — DeepSeek V4 Pro 0813 is the cheaper way to get it. The two are within 0.5 points overall, which is inside the range that effort settings alone move a LiveBench score, so treat them as quality-equivalent. DeepSeek V4 Pro 0813 lists 2.5× cheaper per blended million tokens. When quality ties, cost is the whole decision. DeepSeek V4 Pro 0813 also leads on measured cost per point of capability, at $0.0241 per point.

Is DeepSeek V4 Pro 0813 cheaper than GPT-5.6 Terra?

DeepSeek V4 Pro 0813 is cheaper. On a 3:1 input:output blend, DeepSeek V4 Pro 0813 lists at $1.78 per million tokens and GPT-5.6 Terra at $4.50 — DeepSeek V4 Pro 0813 is 2.5× cheaper. Input and output are priced separately — DeepSeek V4 Pro 0813 charges $1.19 in and $3.56 out, GPT-5.6 Terra charges $2.00 and $12.00 — so the model that looks cheaper flips depending on how output-heavy your workload is.

DeepSeek V4 Pro 0813 vs GPT-5.6 Terra: which scores higher on benchmarks?

DeepSeek V4 Pro 0813 scores 77.4 and GPT-5.6 Terra scores 77.9 overall on LiveBench, the mean of its seven categories. That gap is inside the range that effort settings alone move a score, so treat them as equivalent on published quality. Category scores differ from the overall figure — a model can lead on reasoning and trail on coding, which the per-category table above breaks out.

Which gives better value for money, DeepSeek V4 Pro 0813 or GPT-5.6 Terra?

DeepSeek V4 Pro 0813. Cost per point divides the measured dollars LiveBench spent running the benchmark by the score it earned, so it captures something token pricing misses: a reasoning model can emit many times more tokens than its per-token rate suggests. DeepSeek V4 Pro 0813 works out at $0.0241 per point and GPT-5.6 Terra at $0.1939.

Does DeepSeek V4 Pro 0813 or GPT-5.6 Terra have a bigger context window?

They are effectively the same — 1.0M for DeepSeek V4 Pro 0813 and 1.1M for GPT-5.6 Terra.

Do DeepSeek V4 Pro 0813 and GPT-5.6 Terra support prompt caching?

Both publish a cached-input rate: $0.040 per million for DeepSeek V4 Pro 0813 and $0.200 for GPT-5.6 Terra, against full input rates of $1.19 and $2.00. On a workload with a long stable prefix — a system prompt, a tool schema, a retrieved corpus — that changes the economics more than the headline price does.

Related comparisons

How these numbers are produced

  • Price — provider list price from OpenRouter, refreshed every 15 minutes. “Blended” is a 3:1 input:output mix.
  • ScoresLiveBench release 2026-06-25, using their own category map. Each model shows its strongest published run. A blank means “not evaluated”, never “bad”.
  • Cost per point — the measured dollars LiveBench spent on the run, divided by the score it earned.
  • “Win” — awarded only past a threshold: one full point on a benchmark score, 10% on a price, 25% on a context window. Anything tighter reports as a tie, because effort settings alone move a LiveBench score by more than that.

Published benchmarks rank models on someone else's tasks. Before committing, see LLM & agent evaluation for building an eval on your own.