Gemini 3.6 Flash is a fast, lower-cost legal-work candidate that legal benchmarks have not yet scored
Gemini 3.6 Flash launched on July 21, 2026 as Google's workhorse model, with lower output pricing than Gemini 3.5 Flash and a focus on faster, more efficient work, according to Google's July 21, 2026 announcement. Artificial Analysis measured Gemini 3.6 Flash at 303.6 output tokens per second, the fastest output speed of any model Artificial Analysis tracks, according to the Artificial Analysis model page reviewed July 21, 2026.
No legal-specific benchmark has scored Gemini 3.6 Flash yet: the model has no row on Vals AI's Harvey benchmark, Legal Research Bench, or LegalBench, and no row on Artificial Analysis's Harvey LAB-AA, according to the Vals AI Harvey results reviewed July 21, 2026, Vals AI Legal Research Bench reviewed July 21, 2026, Vals AI LegalBench reviewed July 21, 2026, and Artificial Analysis Harvey LAB-AA results reviewed July 21, 2026. Gemini 3.6 Flash is therefore a strong candidate for high-volume, cost-sensitive legal workflows, and the first direct legal-benchmark results will show how far that case extends to complex agentic legal assignments.
Gemini 3.6 Flash has no direct score on realistic legal assignments
Vals AI's Harvey benchmark tests realistic legal assignments such as memos and redlines rather than short legal questions, according to the Vals AI Harvey benchmark reviewed July 21, 2026. Gemini 3.6 Flash has no Harvey result, while Gemini 3.5 Flash ranks 15th of 24 models by criteria-pass rate with 79.3% criteria-pass and 2.5% task-pass, according to the Vals AI Harvey results reviewed July 21, 2026.
Vals AI's Legal Research Bench measures performance on legal-research work, and Gemini 3.6 Flash has no result while Gemini 3.5 Flash scores 30.77% and sits outside the top 10, according to the Vals AI Legal Research Bench reviewed July 21, 2026. Artificial Analysis's Harvey LAB-AA also tests realistic legal assignments and has no Gemini 3.6 Flash entry, according to the Artificial Analysis Harvey LAB-AA results reviewed July 21, 2026.
The predecessor results measure a different model and do not score Gemini 3.6 Flash. Google reports that Gemini 3.6 Flash improves on the agentic behaviors these benchmarks stress, taking fewer reasoning steps and tool calls in multi-step workflows, according to Google's July 21, 2026 announcement. The most useful next evidence would be a Gemini 3.6 Flash row on Vals AI's Harvey benchmark, Legal Research Bench, and LegalBench, plus Artificial Analysis's Harvey LAB-AA.
Gemini legal results are stronger on short-form reasoning than on agentic legal work
The available benchmark record gives legal teams useful context without directly measuring Gemini 3.6 Flash.
| Benchmark | Available Gemini evidence | What the evidence suggests |
|---|---|---|
| Vals AI Harvey benchmark | Gemini 3.6 Flash has no result; Gemini 3.5 Flash records 79.3% criteria-pass and 2.5% task-pass, ranking 15th of 24 models by criteria-pass rate, according to the Vals AI results reviewed July 21, 2026. | The Flash predecessor sat mid-pack on realistic legal assignments such as memos and redlines. |
| Vals AI Legal Research Bench | Gemini 3.6 Flash has no result; Gemini 3.5 Flash scores 30.77% and sits outside the top 10, while Claude Fable 5 leads at 49.52%, according to the Vals AI results reviewed July 21, 2026. | Open-ended legal research is the area with the most room for the new model to move up. |
| Vals AI LegalBench | Gemini 3.6 Flash has no result; Gemini 3.1 Pro Preview ranks second at 87.40%, Gemini 3 Pro ranks third at 87.02%, and Gemini 3 Flash from December 2025 ranks fifth at 86.86%, according to the Vals AI results reviewed July 21, 2026. | The broader Gemini line performs strongly on shorter legal-reasoning tasks. |
| Artificial Analysis Harvey LAB-AA | Gemini 3.6 Flash has no entry, according to the Artificial Analysis results reviewed July 21, 2026. | Independent scoring of the model's finished legal deliverables has not yet been published. |
The broader Gemini line holds three of the top five positions on Vals AI LegalBench, while the Flash predecessor trails the leaders on Vals AI's agentic Harvey and Legal Research benchmarks, according to the Vals AI LegalBench results reviewed July 21, 2026, Vals AI Harvey results reviewed July 21, 2026, and Vals AI Legal Research Bench reviewed July 21, 2026. LegalBench-style short-form reasoning is encouraging family-level context, and a direct Gemini 3.6 Flash measurement on long, tool-using legal workflows is the piece still to come.
Gemini 3.6 Flash makes speed, context, and token efficiency the practical case
Gemini 3.6 Flash scores 50 on the general, non-legal Artificial Analysis Intelligence Index, well above the tracked-model average of 31, while its measured output speed of 303.6 tokens per second is the fastest Artificial Analysis tracks, according to the Artificial Analysis model page reviewed July 21, 2026. Google prices Gemini 3.6 Flash at $1.50 per million input tokens and $7.50 per million output tokens, according to Google's July 21, 2026 announcement, down from $9.00 per million output tokens for Gemini 3.5 Flash, according to Google's Gemini API pricing reviewed July 21, 2026.
Gemini 3.6 Flash accepts up to 1,048,576 input tokens and produces up to 65,536 output tokens, according to Google's Gemini 3.6 Flash documentation reviewed July 21, 2026. Google reports that Gemini 3.6 Flash uses 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index and up to 65% fewer on DeepSWE, with fewer reasoning steps and tool calls in multi-step workflows, according to Google's July 21, 2026 announcement. Google's efficiency figures are vendor-reported, but the published pricing and context limits make a concrete case for testing the model on long-document and high-volume work.
Gemini 3.6 Flash is available only through Google's hosted products and API
Gemini 3.6 Flash is a closed-weight Google model available through the Gemini API in Google AI Studio and Android Studio, Google Antigravity, Gemini Enterprise Agent Platform, the Gemini Enterprise app, and the Gemini app, according to Google's July 21, 2026 announcement. The stable API model ID is gemini-3.6-flash, according to Google's Gemini 3.6 Flash documentation reviewed July 21, 2026.
Gemini 3.6 Flash supports thinking, function calling, code execution, computer use in preview, search grounding, structured outputs, caching, file search, Google Maps grounding, and URL context, according to Google's Gemini 3.6 Flash documentation reviewed July 21, 2026. Gemini 3.6 Flash does not support audio generation, image generation, or the Live API, according to Google's Gemini 3.6 Flash documentation reviewed July 21, 2026. Hosted-only access matters for legal teams that require on-premises deployment or control over the complete inference environment.
Gemini 3.6 Flash still needs current legal knowledge and careful review
Gemini 3.6 Flash's speed and pricing do not answer whether the model can reliably produce a finished memo, redline, or research product. Legal teams should evaluate the model together with the legal knowledge, instructions, tools, and quality controls surrounding the model.
OpenAgreements offers a free source of up-to-date legal knowledge for areas of law that change quickly. Start with the most-used practice guides, then turn recurring work into a more reliable workflow with legal checklists. Better legal inputs cannot replace professional review, but better legal inputs can help a capable model work from current law instead of generic memory.
Disclosure: OpenAgreements is run by OM Innovation AI Inc., which participates in the Google for Startups program. Google had no role in the research, drafting, or review of this report, and every claim above traces to the dated public sources linked in place.