#104 Claude Sonnet 4.6

Anthropic · Claude Sonnet 4.6

Released: 17.02.2026 · Source

Text; Coding; Reasoning

Use cases: Coding, Reasoning, Text
Accepts: No verified data
Produces: No verified data

Access

Price USD

Provider	Price USD	Conditions
Anthropic · API	$0.3 / 1M cache-read tokens	Standard · Direct Claude API; base input, 5m/1h cache writes, cache hits and output have separate rates. Not a Claude/Claude Code subscription price.Source · 19.09.2026
Anthropic · API	$3.75 / 1M cache-write tokens	Standard · Direct Claude API; base input, 5m/1h cache writes, cache hits and output have separate rates. Not a Claude/Claude Code subscription price.Source · 19.09.2026
Anthropic · API	$3 / 1M input tokens	Standard · Direct Claude API; base input, 5m/1h cache writes, cache hits and output have separate rates. Not a Claude/Claude Code subscription price.Source · 19.09.2026
Anthropic · API	$15 / 1M output tokens	Standard · Direct Claude API; base input, 5m/1h cache writes, cache hits and output have separate rates. Not a Claude/Claude Code subscription price.Source · 19.09.2026

Evaluator & score

Epoch AI · ECI152.25

Composite index · published model-group index · 20.09.2026 · Higher is better

Snapshot: 2026-09-20 · 90% CI [149.69, 154.47]

Official Epoch AI index; combines own, third-party and developer-reported results. It is not a safety audit or a media-generation score.

Conditions

Epoch Capabilities Index; official model-group estimates

Official ECI model scores, Epoch AI, CC BY 4.0. Composite of own, third-party and developer-reported results. 90% bootstrap interval. Not a per-effort score; export date is the model release date, not measurement date. Methodology: https://epoch.ai/eci#documentation

Epoch AI · Chess Puzzles5 %

Independent run · high · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Chess Puzzles; Epoch AI own run; best score across scorers

Epoch AI own run; Chess Puzzles; version: claude-sonnet-4-6_high; mode: high; source id: 5Q9RNC2JfEmmvNFFE5CQCf; file: chess_puzzles.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 13.07.2026

Epoch AI · Chess Puzzles3 %

Independent run · max · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Chess Puzzles; Epoch AI own run; best score across scorers

Epoch AI own run; Chess Puzzles; version: claude-sonnet-4-6_max; mode: max; source id: TDT98jZXHbUMFYTphGEpqC; file: chess_puzzles.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 06.08.2026

Epoch AI · Chess Puzzles8 %

Independent run · medium · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Chess Puzzles; Epoch AI own run; best score across scorers

Epoch AI own run; Chess Puzzles; version: claude-sonnet-4-6_medium; mode: medium; source id: eJBzZyT5ohb8ieaJCum5rK; file: chess_puzzles.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 13.07.2026

Epoch AI · Chess Puzzles13 %

Independent run · thinking budget 32K · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Chess Puzzles; Epoch AI own run; best score across scorers

Epoch AI own run; Chess Puzzles; version: claude-sonnet-4-6_32K; mode: thinking budget 32K; source id: N5beL4W3WahAubJgDKivnE; file: chess_puzzles.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 20.02.2026

Epoch AI · FrontierMath32.4 %

Independent run · thinking budget 16K · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

FrontierMath; Epoch AI own run; best score across scorers

Epoch AI own run; FrontierMath-2025-02-28-Private; version: claude-sonnet-4-6_16K; mode: thinking budget 16K; source id: ewFjUhQhYidjkErDdLvcQL; file: frontiermath.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 20.02.2026

Epoch AI · FrontierMath-Tier-4-2025-07-01-Private8.3 %

Independent run · thinking budget 16K · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

FrontierMath-Tier-4-2025-07-01-Private; Epoch AI own run; best score across scorers

Epoch AI own run; FrontierMath-Tier-4-2025-07-01-Private; version: claude-sonnet-4-6_16K; mode: thinking budget 16K; source id: RTfHBvZX3Bh5EPPj6Cf3ab; file: frontiermath_tier_4.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 21.02.2026

Epoch AI · GPQA Diamond83.333 %

Independent run · high · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

GPQA Diamond; Epoch AI own run; best score across scorers

Epoch AI own run; GPQA diamond; version: claude-sonnet-4-6_high; mode: high; source id: DmdKPD9Z9RXwYyATAoG4Uw; file: gpqa_diamond.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 13.07.2026

Epoch AI · GPQA Diamond78.788 %

Independent run · max · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

GPQA Diamond; Epoch AI own run; best score across scorers

Epoch AI own run; GPQA diamond; version: claude-sonnet-4-6_max; mode: max; source id: SEePRKRcv8oqn5HpD2KFxB; file: gpqa_diamond.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 06.08.2026

Epoch AI · GPQA Diamond83.333 %

Independent run · medium · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

GPQA Diamond; Epoch AI own run; best score across scorers

Epoch AI own run; GPQA diamond; version: claude-sonnet-4-6_medium; mode: medium; source id: e8MXnAM8HMF8eEzYtYA7xv; file: gpqa_diamond.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 13.07.2026

Epoch AI · GPQA Diamond87.374 %

Independent run · thinking budget 32K · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

GPQA Diamond; Epoch AI own run; best score across scorers

Epoch AI own run; GPQA diamond; version: claude-sonnet-4-6_32K; mode: thinking budget 32K; source id: 58eQmyCfPa3FufhXJFvAL2; file: gpqa_diamond.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 20.02.2026

Epoch AI · Mock AIME 2024/202575.556 %

Independent run · high · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Mock AIME 2024/2025; Epoch AI own run; best score across scorers

Epoch AI own run; OTIS Mock AIME 2024-2025; version: claude-sonnet-4-6_high; mode: high; source id: oXZ7ZVZtADrwFPKAe6zTvL; file: otis_mock_aime_2024_2025.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 13.07.2026

Epoch AI · Mock AIME 2024/202571.111 %

Independent run · max · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Mock AIME 2024/2025; Epoch AI own run; best score across scorers

Epoch AI own run; OTIS Mock AIME 2024-2025; version: claude-sonnet-4-6_max; mode: max; source id: XnQVH5qSfxVjpzoPwUEmnh; file: otis_mock_aime_2024_2025.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 06.08.2026

Epoch AI · Mock AIME 2024/202582.222 %

Independent run · medium · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Mock AIME 2024/2025; Epoch AI own run; best score across scorers

Epoch AI own run; OTIS Mock AIME 2024-2025; version: claude-sonnet-4-6_medium; mode: medium; source id: kDNNbUYXhrAH5uDBGykFaf; file: otis_mock_aime_2024_2025.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 13.07.2026

Epoch AI · Mock AIME 2024/202585.794 %

Independent run · thinking budget 32K · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Mock AIME 2024/2025; Epoch AI own run; best score across scorers

Epoch AI own run; OTIS Mock AIME 2024-2025; version: claude-sonnet-4-6_32K; mode: thinking budget 32K; source id: 9LWdzVN5w4ihNvncsaCWMu; file: otis_mock_aime_2024_2025.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 20.02.2026

Epoch AI · Mystery Game Puzzles14 %

Independent run · default (effort unspecified by source) · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Mystery Game Puzzles; Epoch AI own run; best score across scorers

Epoch AI own run; Mystery Game Puzzles; version: claude-sonnet-4-6; mode: default (effort unspecified by source); source id: VjfWxdhdNerrh4huSHT8PT; file: mystery_game_puzzles.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 06.08.2026

Epoch AI · Mystery Game Puzzles16 %

Independent run · low · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

Mystery Game Puzzles; Epoch AI own run; best score across scorers

Epoch AI own run; Mystery Game Puzzles; version: claude-sonnet-4-6_low; mode: low; source id: YdyvoMQ4mA5V8Ymg6Tksxu; file: mystery_game_puzzles.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 06.08.2026

Epoch AI · SWE-bench Verified75.207 %

Independent run · default (effort unspecified by source) · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

SWE-bench Verified; Epoch AI own run; best score across scorers

Epoch AI own run; source version: claude-sonnet-4-6; file: swe_bench_verified.csv; row: KEdgiMmtJKHxgGiS82agYj.

Measurement date: 21.02.2026

Epoch AI · SimpleQA Verified35.5 %

Independent run · high · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

SimpleQA Verified; Epoch AI own run; best score across scorers

Epoch AI own run; SimpleQA Verified; version: claude-sonnet-4-6_high; mode: high; source id: LJyKf8Xy4dymxNHbGLwNdA; file: simpleqa_verified.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 10.08.2026

Epoch AI · SimpleQA Verified32.8 %

Independent run · max · 20.09.2026 · Higher is better

Snapshot: 2026-09-20

Conditions

SimpleQA Verified; Epoch AI own run; best score across scorers

Epoch AI own run; SimpleQA Verified; version: claude-sonnet-4-6_max; mode: max; source id: Zs5fCvNzaz8LhqgwDSAztU; file: simpleqa_verified.csv. CC BY 4.0, Epoch AI. Snapshot 2026-09-20. Source column Best score (across scorers), not AIpedia's selection of the best effort.

Measurement date: 27.08.2026

SourceVerified 18.09.2026