// head_to_head

Claude Sonnet 4.6 vs GPT-5.2

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

GPT-5.2 wins outright — it scores higher and costs less.

GPT-5.2 leads by 1.6 points overall while listing 25% cheaper per blended million tokens. There is no trade-off to reason about here; the only reason to pick the other is a constraint this table does not show, like an existing contract, a region, or a provider you are already on. GPT-5.2 also leads on measured cost per point of capability, at $0.1289 per point.

anthropic

Claude Sonnet 4.6

Blended / 1M
$6.00
Context
1M
Released
Feb 17, 2026
Overall score
73.0
reasoningtool callingimage inputfile inputprompt caching

openai

GPT-5.2

Blended / 1M
$4.81
Context
400K
Released
Dec 10, 2025
Overall score
74.6
reasoningtool callingfile inputimage inputprompt caching

Specs and pricing

MetricClaude Sonnet 4.6GPT-5.2
LiveBench overall

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

73.074.6win
Cost per point

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

$0.1561$0.1289win
Blended price / 1M

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

$6.00$4.81win
Input price / 1M$3.00$1.75win
Output price / 1M$15.00$14.00
Cached input / 1M

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

$0.300$0.175win
Context window1Mwin400K
Max output tokens128K128K

Benchmarks by category

An overall score averages away the thing you probably care about. These are the seven LiveBench categories scored separately — Claude Sonnet 4.6 on top, GPT-5.2 below, both out of 100.

Agentic coding
42.6
50.3
Coding
79.3
76.1
Reasoning
84.8
83.2
Mathematics
87.0
93.2
Data analysistoo close to call
77.9
78.2
Language
76.1
79.8
Instruction following
63.2
61.8

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.

WorkloadClaude Sonnet 4.6GPT-5.2
Support chatbot

1.2K in / 400 out × 200K requests

$1725.60/mo$1426.60/mo
RAG assistant

8K in / 600 out × 100K requests

$2220.00/mo$1610.00/mo
Coding agent

40K in / 4K out × 20K requests

$2088.00/mo$1638.00/mo
Document extraction

20K in / 1.5K out × 50K requests

$3990.00/mo$2721.25/mo
Bulk classification

500 in / 20 out × 5M requests

$7650.00/mo$4987.50/mo
Run these two through the cost calculator

Which should you pick?

You are running this at volume

GPT-5.2

Lowest measured cost per point of capability at $0.1289 per point — the gap compounds with every request.

Quality matters more than the bill

GPT-5.2

Highest overall LiveBench score of the two at 74.6.

The workload is coding or agentic work

GPT-5.2

Leads on agentic coding — 50.3 against 42.6.

You need to fit large documents in one call

Claude Sonnet 4.6

Wider context window — 1M against 400K.

Claude Sonnet 4.6 vs GPT-5.2 FAQ

Which is better, Claude Sonnet 4.6 or GPT-5.2?

GPT-5.2 wins outright — it scores higher and costs less. GPT-5.2 leads by 1.6 points overall while listing 25% cheaper per blended million tokens. There is no trade-off to reason about here; the only reason to pick the other is a constraint this table does not show, like an existing contract, a region, or a provider you are already on. GPT-5.2 also leads on measured cost per point of capability, at $0.1289 per point.

Is Claude Sonnet 4.6 cheaper than GPT-5.2?

GPT-5.2 is cheaper. On a 3:1 input:output blend, Claude Sonnet 4.6 lists at $6.00 per million tokens and GPT-5.2 at $4.81 — GPT-5.2 is 25% cheaper. Input and output are priced separately — Claude Sonnet 4.6 charges $3.00 in and $15.00 out, GPT-5.2 charges $1.75 and $14.00 — so the model that looks cheaper flips depending on how output-heavy your workload is.

Claude Sonnet 4.6 vs GPT-5.2: which scores higher on benchmarks?

Claude Sonnet 4.6 scores 73.0 and GPT-5.2 scores 74.6 overall on LiveBench, the mean of its seven categories. That is a 1.6-point lead for GPT-5.2. 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, Claude Sonnet 4.6 or GPT-5.2?

GPT-5.2. 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. Claude Sonnet 4.6 works out at $0.1561 per point and GPT-5.2 at $0.1289.

Does Claude Sonnet 4.6 or GPT-5.2 have a bigger context window?

Claude Sonnet 4.6 has the larger context window: 1M for Claude Sonnet 4.6 against 400K for GPT-5.2. Note that a window you can fill is not a window you should fill — retrieval quality usually degrades well before the limit, and you pay for every token you put in it.

Do Claude Sonnet 4.6 and GPT-5.2 support prompt caching?

Both publish a cached-input rate: $0.300 per million for Claude Sonnet 4.6 and $0.175 for GPT-5.2, against full input rates of $3.00 and $1.75. 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.