Models › Deepseek V4.1
DeepSeek • Launched September 2026 with native vision
DeepSeek V4.1 Flash is DeepSeek’s open-weight model for agentic work
Starting provider API rate; context tiers apply
See what V4.1 Flash does best, then review its limits, benchmarks, and pricing below
GPT-5.6 Sol is the flagship model in OpenAI’s GPT-5.6 series.
Every V4.1 Flash result below comes from a named source and is labeled by protocol, so each score reflects how it was measured
| Benchmark | Score | Evaluation details | Source |
|---|---|---|---|
Humanity's Last Exam A 2,500-question expert-level benchmark spanning dozens of academic fields. Tool-assisted and no-tools results are separate protocols. |
36.8% Accuracy |
Version: Humanity's Last Exam, full set · Protocol: No external tools · Harness: Artificial Analysis independent evaluation · Reasoning effort: max | Methodology Artificial Analysis Independently reproduced |
SciCode A benchmark of realistic scientific-computing problems drawn from published research, testing whether generated code produces correct results. |
51.9% Accuracy |
Version: SciCode · Harness: Artificial Analysis independent evaluation · Reasoning effort: max | Methodology Artificial Analysis Independently reproduced |
AA-LCR (Long-Context Recall) A long-context recall benchmark testing whether a model can retrieve and reason over specific facts buried in a long document. |
84.0% Accuracy |
Version: AA-LCR · Harness: Artificial Analysis independent evaluation · Reasoning effort: max | Methodology Artificial Analysis Independently reproduced |
Terminal-Bench 2.1 A terminal-based agent benchmark of 89 tasks, graded on whether an isolated sandbox reaches the correct final state. |
90.6% Pass@1 |
Version: Terminal-Bench 2.1 · Split: 89 terminal tasks · Harness: DeepSeek internal evaluation · Reasoning effort: max | Methodology DeepSeek Provider-reported |
DeepSWE 1.1 A long-horizon software-engineering benchmark with 113 original tasks graded by hand-written tests. |
74.2% Pass@1 (Resolved) |
Version: DeepSWE 1.1 · Split: 113 tasks across 91 repositories · Harness: DeepSeek internal evaluation · Reasoning effort: max | Methodology DeepSeek Provider-reported |
Arena rankings capture how people vote between anonymous responses. They are reported separately from task benchmarks and are not used as substitutes for missing benchmark results
No exact Arena match is available for this model. Similar names and provider variants are not merged automatically.
DeepSeek V4.1 Flash runs a 1,048,576-token context window with image input, open weights, and adjustable reasoning, priced on a peak/off-peak schedule below
| Model | |
| Provider | DeepSeek |
| Release date | Sep 10, 2026 |
| Catalog added | Sep 22, 2026 |
| Context window | 1,048,576 tokens |
| Maximum output | 384,000 tokens |
| Knowledge cutoff | Not available in reviewed sources |
| License | MIT License (open-weight) |
| Capabilities and API | |
| Parameters | 552B backbone • 8B active (prefill) / 16B active (decode) |
| Input types | Image, Text |
| Output types | Text |
| Reasoning | Supported (thinking/non-thinking modes, adjustable effort) |
| Provider endpoint tool parameters | Accepts tool parameters |
| Off-peak API input (cache miss / cache hit) | $0.15 / 1M tokens • $0.003 / 1M tokens |
| Off-peak API output | $0.60 / 1M tokens |
| Peak API input (cache miss / cache hit) | $0.30 / 1M tokens • $0.006 / 1M tokens |
| Peak API output | $1.20 / 1M tokens |
Provider token rates are context-dependent. The base rates apply below the listed prompt thresholds; the matching tier applies at or above each threshold. These are reference provider API rates, not Pyxa AI checkout charges — access in Pyxa AI follows the allowances of your Pyxa AI plan.
Each V4.1 Flash benchmark and specification links to its original source, so you can check where every figure comes from
Open another model’s detail page or compare it directly with DeepSeek V4.1 Flash. Suggestions come from the current catalog
Put DeepSeek V4.1 Flash to work on your coding, research, and long-context tasks inside Pyxa AI