DeepSeek, released Dec 1, 2025
DeepSeek V3.2price, context, benchmarks and release details
Provisional: not enough results to rank yet Open weights 56% confidence 56 percent, Medium confidence, 3 of 8 expected sources in
53.3
SI Score
#96 of 142 ranked models
- Input, per 1M tokens
- $0.28LiteLLMFirst-party API Standard token rate; MIT LiteLLM transcription. Provider documentation: not supplied in the MIT entry; rate is a transcription, not independently verified. Exact endpoint only; cache/batch/long-context rates excluded
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Output, per 1M tokens
- $0.40LiteLLMFirst-party API Standard token rate; MIT LiteLLM transcription. Provider documentation: not supplied in the MIT entry; rate is a transcription, not independently verified. Exact endpoint only; cache/batch/long-context rates excluded
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Context window
- 128Kmodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Max output
- 64Kmodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Released
- Dec 1, 2025models.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗
How the score breaks down
Coding (weight 40 percent) 70.0
Math (weight 15 percent) —
Preference (weight 15 percent) 75.0
Reasoning (weight 30 percent) 31.6
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
- 94 Claude Sonnet 4.5 53.9
- 95 Inkling Small 53.6
- 96 DeepSeek V3.2 53.3
- 97 GLM-4.7-Flash 53.3
- 98 Step 3.5 Flash 53.0
Benchmark results
6 benchmarks, 6 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.
ARC-AGI-1 (public eval)reasoning 61.6%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] deepseek-v3.2Published Oct 6, 2026
Retrieved Oct 9, 2026 · factual citation
Open source ↗ 61.6
Open source ↗ 61.6
ARC-AGI-1 (semi-private)reasoning 57.0%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] deepseek-v3.2Published Oct 6, 2026
Retrieved Oct 9, 2026 · factual citation
Open source ↗ 57.0
Open source ↗ 57.0
ARC-AGI-2 (public eval)reasoning 3.9%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] deepseek-v3.2Published Oct 6, 2026
Retrieved Oct 9, 2026 · factual citation
Open source ↗ 3.9
Open source ↗ 3.9
ARC-AGI-2 (semi-private)reasoning 4.0%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] deepseek-v3.2Published Oct 6, 2026
Retrieved Oct 9, 2026 · factual citation
Open source ↗ 4.0
Open source ↗ 4.0
SWE-bench Verifiedcoding 70%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] mini-SWE-agent; high; 2.0.0Published Feb 17, 2026
Retrieved Oct 9, 2026 · factual citation
Open source ↗ 70.0
Open source ↗ 70.0
LMArena Textpreference 1420.0 eloLMArena / ArenaPublished source fact [variant] text / overallPublished Oct 8, 2026
Retrieved Oct 9, 2026 · CC-BY-4.0
Open source ↗ 75.0
Open source ↗ 75.0
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
- Yesmodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - License
- MIT Licensemodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - Input modalities
- textmodels.devPublished source fact
Retrieved Oct 9, 2026 · MIT
Open source ↗ - First seen by SuperIndex
- Oct 8, 2026
- Coverage
- 45% of expected source weight
Reported (3)
- ARC PrizeOct 8, 2026
- LMArena / ArenaOct 8, 2026
- SWE-bench VerifiedOct 8, 2026
Awaiting (5)
- Epoch AI Benchmarking23% of weight
- Humanity’s Last Exam11% of weight
- Official model cards via models.dev4% of weight
- LiveBench11% 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.