VerdictPal · editorial desk · 2026VerdictPal

Compare · Two models, head-to-head

Pick two models. See the diff.

Composite scores, benchmark ranks, capability grids, and pricing per million tokens. All sourced from the same data that powers the model atlas.

Cards Models

Claude Sonnet 4.6 vs GPT 5.4

Model A

Model B

MODEL A

Claude Sonnet 4.6

Anthropic · Claude Sonnet

Composite 66Intel 6024 rows11 sources

Anthropic's February 2026 mid-tier model. Vals Index 60.30% (#3 on the public leaderboard — beats GPT 5.4 Mini and Gemini 3 Flash). SWE-bench Verified 79.6%, ARC-AGI-2 58.3%. 1M context, $3 / $15 per 1M tokens — 5× cheaper than Opus 4.8 on input.

MODEL B

GPT 5.4

OpenAI · GPT 5

Composite 61Intel 5118 rows10 sources

OpenAI's March 2026 model. AA Intelligence Index 56.8 (#5 globally). SWE-bench Verified 80.0%, Terminal-Bench 2.0 81.8% (ForgeCode). ARC-AGI-2 83.3% (Pro variant). 256K context, $7 / $28 per 1M tokens — 44% cheaper than GPT 5.5 on input.

Claude Sonnet 4.66stats won

Tied0stats

GPT 5.41stats won

Spec showdown

Capacity, price, freshness.

Claude Sonnet 4.6vsGPT 5.4

Composite scores are within 5 points — 66 for Claude Sonnet 4.6 vs 61 for GPT 5.4. The choice comes down to capability profile and price. Claude Sonnet 4.6 has the higher Vals/AA intelligence score (60 vs 51). Claude Sonnet 4.6 is more affordable at $3.0/1M input — about 2× cheaper than $7.0/1M. Claude Sonnet 4.6's 1M context window dwarfs GPT 5.4's 256K — a meaningful difference for long-document or codebase work.

Composite score
Claude Sonnet 4.666Winner on this rowGPT 5.461
Intelligence (Vals/AA)
Claude Sonnet 4.660Winner on this rowGPT 5.451
Context window
Claude Sonnet 4.61MWinner on this rowGPT 5.4256K
Max output
Claude Sonnet 4.6128KWinner on this rowGPT 5.466K
Input price / 1M (lower wins)
Claude Sonnet 4.6$3.0Winner on this rowGPT 5.4$7.0
Output speed (AA default API)
Claude Sonnet 4.660 t/sGPT 5.4152 t/sWinner on this row
Time to first token (AA default API)
Claude Sonnet 4.61.18sWinner on this rowGPT 5.4129.43s

Composite score

Intelligence (Vals/AA)

Context window

256K

Max output

128K

66K

Input price / 1M (lower wins)

$3.0

$7.0

Output speed (AA default API)

60 t/s

152 t/s

Time to first token (AA default API)

1.18s

129.43s

Capability grid

What each model can do.

Modalities

Claude Sonnet 4.6GPT 5.4

Text●●

Image—●

Audio——

Video——

Code●●

Tool use●●

Reasoning●●

Access

Claude Sonnet 4.6GPT 5.4

API●●

Consumer app●●

Open weights——

Managed cloud●●

On-device——

Family radar

Top score per benchmark family.

Knowledge

Reasoning

Coding

Agentic

Long context

Tool use

Human preference

Benchmark ranks

Every shared benchmark, side by side.

Knowledge

MMLU#4/22 · #7/22

90.6%

89.8%

Reasoning

ARC-AGI#8/14 · #14/14

58.3%

0.2%

Artificial Analysis Intelligence Index#43/67 · #11/67

35.9%

51.4%

GPQA Diamond#54/69 · #13/69

77.2%

92%

Humanity's Last Exam#54/65 · #10/65

13.2%

41.6%

IFBench#54/57 · #22/57

41.16%

73.95%

LiveBench#5/21 · #6/21

67.2%

66.4%

Coding

SciCode#37/65 · #5/65

46.9%

56.6%

SWE-bench Verified#13/23 · #12/23

79.6%

80%

Agentic

Terminal-Bench#24/69 · #7/69

68.4%

81.8%

τ³-Bench Banking#12/64 · #26/64

79.53%

30.31%

Long context

AA-LCR#51/64 · #3/64

57.67%

74%

LongBench v2#8/19 · #6/19

52.6%

54.2%

Tool use

BFCL#1/20 · #5/20

88.7%

83.8%

Human preference

LMArena (Chatbot Arena)#10/13 · #8/13

1396

1405

Performance

AA output speed#48/69 · #17/69

60 t/s

152 t/s

AA time to first token#39/69 · #1/69

1.18s

129.43s

Dossier facts

Dates, sources, fineprint.

Claude Sonnet 4.6GPT 5.4

ProviderAnthropicOpenAI

FamilyClaude SonnetGPT 5

Release date2026-02-172026-03-05

Knowledge cutoff2025-12-012026-01-31

Context window1M256K

Max output128K66K

Input price /1M$3.0$7.0

Output price /1M$15$28

Statusflagshipflagship

Composite score6661

Intelligence score6051

Intel sourcevalsaa

Benchmark rows2418

Source count1110

Freshnessfreshfresh

Pricing checked2026-06-012026-06-01

Ecosystem

Which tools wrap each model.

Claude Sonnet 4.6

GPT 5.4

Verdict

Which model to actually pick.

Open Claude Sonnet 4.6's dossier Open GPT 5.4's dossier