Privaj Scout lists Deep Research Preview (Apr-21-2026) as a Google image-capable model with a 131.07K-token context window, provider-terms licensing, and multimodal inputs (text, image, video, audio, PDF) producing text and image outputs. Published capability labels cover vision, video understanding, image generation, The page is a third-party capability aggregator that explicitly states no benchmark results have been imported and disclaims independent evaluation, so it does not provide performance or behavioral evidence about the model. It confirms the exact model slug, modality set, context size, and first-party pricing but adds n
Model details
Deep Research Preview (Apr-21-2026)
Deep Research Preview sits inside Google's gemini-pro lineup as an experimental endpoint oriented around agentic, multi-step research workflows. Rather than a single-turn chat model, it is positioned as a hosted research agent that can ingest a broad mix of inputs and orchestrate tool use, citations, and extended reasoning to produce synthesized, long-form reports. Its place in the gemini-pro family signals a deliberate tie-in with Google's larger Gemini roadmap, suggesting that improvements in the underlying Gemini reasoning stack will flow directly into this preview as it matures toward general availability.
In practical terms, this preview is aimed at builders who need deep, sourced investigations rather than quick answers: analysts assembling market or scientific briefs, product researchers mapping competitive landscapes, and teams building knowledge pipelines where provenance, tool integration, and temperature-tunable reasoning matter. The supported evidence shows a flat base pricing structure of $2.00 per million input tokens and $12.00 per million output tokens, with cache reads priced at $0.20 per million tokens, placing it at the higher end of flagship-tier rates and signaling a premium, agentic use case. For teams that prioritize long-context, multi-modal research outputs and are comfortable paying flagship rates for a Gemini Pro-tier engine, the model offers a compelling fit; lighter, latency-sensitive workloads are better served by smaller, cheaper Gemini variants.
Quick Info
Powered by- Provider
- Model key
- deep-research-preview-04-2026
- Release date
- Apr 21, 2026
- Last updated
- Apr 21, 2026
- Knowledge cutoff
- 2025-01
- Input modalities
- Output modalities
- Capabilities
Cost
- Input token cost
- $2.00
- Output token cost
- $12.00
Limits
- Output tokens
- 65,536 tokens
- Context window
- 131,072 tokens
Latest news about Deep Research Preview (Apr-21-2026)
Llmpricing.dev's page for deep-research-preview-04-2026 reports Google first-party pricing of $2.00 input / $12.00 output per million tokens with cache read at $0.20, a 131,072-token context window, a 65,536-token output limit, a 2025-01 knowledge cutoff, and a 2026-04-21 release/updated date in the gemini-pro series. Like the other aggregator candidates, this is a third-party pricing/spec page that restates Google vendor data rather than reporting a release or capability change. It confirms the exact model ID and a few useful details (200K-tier input rate, cache-read pricing) but provides no benchmarks, release notes, or behavioral
Modelbenchmark.io lists Deep Research Preview (Apr-21-2026) as a Google agentic model for autonomous multi-step research, synthesis, and cited reports. It carries a 131K-token context window, 66K-token max output, a 2025-01 knowledge cutoff, a 2026-04-21 release, and closed weights, with multimodal inputs (text, image, The page is a third-party aggregator mirroring Google's vendor spec sheet rather than an official announcement, so it does not provide benchmarks, behavioral notes, or release commentary. It does consistently corroborate the exact model slug, context length, output limit, multimodal modality set, and pricing structure