Anthropic, released Sep 29, 2025
Claude Sonnet 4.5price, context, benchmarks and release details
Provisional: not enough results to rank yet 80% confidence 80 percent, High confidence, 3 of 7 expected sources in
53.9
SI Score
#94 of 142 ranked models
- Input, per 1M tokens
- $3.00models.devFirst-party hosted API; MIT models.dev transcription. Provider documentation: https://docs.anthropic.com/en/docs/about-claude/models. Exact canonical endpoint; lowest short-context Standard USD token tier; cache/batch discounts excluded. Deprecated endpoints excluded; anthropic/claude-sonnet-4-5-20250929
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Output, per 1M tokens
- $15.00models.devFirst-party hosted API; MIT models.dev transcription. Provider documentation: https://docs.anthropic.com/en/docs/about-claude/models. Exact canonical endpoint; lowest short-context Standard USD token tier; cache/batch discounts excluded. Deprecated endpoints excluded; anthropic/claude-sonnet-4-5-20250929
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Context window
- 200Kmodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Max output
- 64Kmodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Released
- Sep 29, 2025models.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗
How the score breaks down
Coding (weight 40 percent) 71.3
Math (weight 15 percent) 44.8
Preference (weight 15 percent) 76.4
Reasoning (weight 30 percent) 31.4
Weights: reasoning 30%, math 15%, coding 40%, preference 15%. Results use fixed 0–100 scales before averaging, and thin evidence is pulled toward 50. Method si-v3-retained-evidence-2, computed Oct 9, 2026, 05:13 UTC.
Around it on the leaderboard
- 92 Grok 4.3 54.6
- 93 DeepSeek-V3 54.1
- 94 Claude Sonnet 4.5 53.9
- 95 Inkling Small 53.6
- 96 DeepSeek V3.2 53.3
Benchmark results
8 benchmarks, 14 resultsEach row shows the best published result. Where a model was tested at several settings, such as reasoning effort, open the row to see each one. Hover or tap a value for its source.
GPQA Diamondreasoning 82.3%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 59KPublished Oct 28, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗
82.3 4 settings
- 16K thinking 78.8%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 16KPublished Oct 28, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ - 32K thinking 81.7%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 32KPublished Oct 21, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ - 59K thinking 82.3%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 59KPublished Oct 28, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ - Setting 4 73.7%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Sep 29, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗
Humanity's Last Exam (Scale AI)reasoning 7.5%Humanity’s Last ExamPotential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. [variant] Published Oct 2, 2025
Retrieved Oct 9, 2026 · factual citation
Open source ↗ 7.5
Open source ↗ 7.5
FrontierMath Tier 4 (v2)math 2.4%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 32KPublished Jun 11, 2026
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 2.4
Open source ↗ 2.4
FrontierMath Tiers 1–3 (v2)math 23.9%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 32KPublished Jun 11, 2026
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 23.9
Open source ↗ 23.9
MATH Level 5math 97.7%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 32KPublished Oct 21, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 97.7
Open source ↗ 97.7
OTIS Mock AIME 2024–2025math 77.8%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 32KPublished Oct 21, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗
77.8 4 settings
- 16K thinking 71.1%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 16KPublished Oct 28, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ - 32K thinking 77.8%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 32KPublished Oct 21, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ - 59K thinking 77.8%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] 59KPublished Oct 28, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ - Setting 4 35.6%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Sep 29, 2025
Retrieved Oct 9, 2026 · CC-BY
Open source ↗
SWE-bench Verifiedcoding 71.3%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Feb 5, 2026
Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 71.3
Open source ↗ 71.3
LMArena Textpreference 1434.5 eloLMArena / ArenaPublished source fact [variant] text / overallPublished Oct 8, 2026
Retrieved Oct 9, 2026 · CC-BY-4.0
Open source ↗ 76.4
Open source ↗ 76.4
Normalization uses a fixed 0–100 scale for each unit, independent of other models. Compare evaluation conditions before reading a small gap as decisive. “Lab-reported” marks the provider's own published figure.
Details and sources
- Open weights
- Nomodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - License
- not yet reported
- Input modalities
- text, image, pdfmodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - First seen by SuperIndex
- Oct 8, 2026
- Coverage
- 64% of expected source weight
Reported (3)
- Epoch AI BenchmarkingOct 8, 2026
- Humanity’s Last ExamOct 8, 2026
- LMArena / ArenaOct 8, 2026
Awaiting (4)
- ARC Prize13% of weight
- Official model cards via models.dev4% of weight
- LiveBench13% of weight
- Terminal-Bench6% of weight
Confidence rises as pending sources publish. Some sources never cover some models, so confidence reaches 100% at 80% of expected weight.