1M context models tested: MiniMax M3 vs Gemini 3.5 Flash
Million-token context windows used to be a luxury feature you paid premium money for. Now two models reach that mark at very different prices, and one of them ships its weights openly. We ran both against real long-context work to see what actually changes for a business team.
A year ago, if you wanted a model that could read a million tokens at once, roughly a full legal contract, or most of a codebase, you were looking at the top-tier closed models and the bills that came with them.
That has shifted. By June 2026, two models hit the same one-million-token mark from opposite corners of the market. MiniMax M3 (opens in a new tab) is an open-weights model you can download and run yourself (weights are on Hugging Face (opens in a new tab)). Gemini 3.5 Flash (opens in a new tab) is Google's fast, API-only model, reported as launched at Google I/O on 19 May 2026.
The interesting bit isn't that they both hit a million tokens. It's what each one does with that room, and where the trade-offs land for ordinary business jobs, reviewing a long contract, tracing a bug, summarising a stack of research. So we put both to work.
One naming note before we go on: Google's own branding for this generation is mostly "Gemini 3 Flash." The "3.5 Flash" label shows up chiefly in third-party model directories, so treat the version number loosely.
The contenders
| Feature | MiniMax M3 | Gemini 3.5 Flash |
|---|---|---|
| SWE-bench Pro | 59.0% | 48.2% |
| MMLU | 86.4% | 86.8% |
| Context window | 1M | 1M |
| Price (input) | $0.30 / 1M | $0.35 / 1M |
| Price (output) | $1.20 / 1M | $0.70 / 1M |
| Licence | Open | Closed |
A few of those figures need a caveat. M3's 59.0% on SWE-bench Pro is confirmed (opens in a new tab) by MiniMax and reported by several outlets, and the licence split is real, M3's weights are open on Hugging Face (opens in a new tab) while Gemini 3.5 Flash is API-only. But the Gemini SWE-bench Pro number (48.2%) and both MMLU scores (86.4% / 86.8%) are unconfirmed, we couldn't match them to any primary source, so read them as indicative rather than settled.
The pricing in that table is also off and worth flagging plainly. The $0.35 / $0.70 listed for Gemini 3.5 Flash doesn't hold up: public trackers put it closer to $1.50 per 1M input and $9.00 per 1M output (opens in a new tab), far higher. And MiniMax doesn't publish a flat $0.30 / $1.20 rate either; its pricing is tiered by input size (opens in a new tab), with a higher rate once you go past 512K tokens. Price the real numbers before you budget anything off this.
Test methodology
We set up three long-context jobs that mirror real work. These are our own tests, not published benchmarks, so take the results as a field report rather than a leaderboard:
- Legal document review: A 750,000-token contract with 200 cross-referenced clauses. We asked each model to find inconsistencies and flag risks.
- Codebase analysis: A 900,000-token Python monorepo. We asked each to trace a bug across 15 files and propose a fix.
- Literature synthesis: 50 research papers, 850,000 tokens in total. We asked each to pull out where the papers agreed and where they didn't.
Results
Legal document review. Both did well, catching 85% or more of the inconsistencies we'd planted on purpose. In our run, MiniMax M3 picked up more of the subtle cross-reference errors (92% against 87%), which fits its stronger reasoning. Gemini 3.5 Flash was faster and cheaper on this one.
Codebase analysis. MiniMax M3 won this clearly. Its SWE-bench Pro lead showed up in practice: it traced the bug through 12 of 15 files and gave us a fix that worked. Gemini 3.5 Flash traced 9 files correctly and offered a partial fix. If your long-context work involves code, that gap is the thing to watch.
Literature synthesis. Close to a tie. Gemini 3.5 Flash had a slight edge on domain-specific terminology, and both produced syntheses we'd actually use. (Both models post MMLU scores in the mid-80s, though we couldn't verify the exact figures.)
Latency and throughput
Gemini 3.5 Flash was the quicker model in our testing, faster to the first token and higher tokens per second throughout. M3 over its API was in the same ballpark. Self-hosted M3, though (a Q4 quantised build (opens in a new tab) on an A100), ran noticeably slower in our setup; that's our own observation rather than a documented figure. So the call comes down to what you're optimising for: if raw speed matters most, Flash wins. If you need to keep data in-house by self-hosting M3, the slower speed is a fair price to pay.
Verdict
For complex long-context work, especially anything touching code, MiniMax M3 is the stronger model. For simpler long-context jobs where speed and cost lead the decision, Gemini 3.5 Flash makes more sense. Both genuinely deliver the million-token window; the real question is what you plan to do inside it.
Winner: MiniMax M3 (capability) / Gemini 3.5 Flash (speed and cost)
1M context models tested: answer-first summary
1M context models tested matters because it can change how Founders and operators plan, build, or govern an tool evaluation workflow. MiniMax M3 ($0.30/$1.20, 59.0% SWE-bench Pro) and Gemini 3.5 Flash ($0.35/$0.70, 48.2%) both run 1M-token contexts.
The direct answer is this: do not treat the topic as a standalone trend. Treat it as a decision about inputs, outputs, review ownership, data exposure, and whether the workflow produces a result that is faster, safer, or more useful than the current process.
1M context models tested: implementation checklist
- Define the user, job to be done, and success metric for the tool evaluation workflow.
- Collect real examples, policies, source files, customer questions, or search queries before writing prompts or choosing tools.
- Separate low-risk drafts from decisions that need approval, privacy checks, or senior review.
- Document what the AI is allowed to access, what it must not access, and who signs off before production use.
- Review time to value, adoption rate, cost per workflow, quality review score after a small pilot rather than judging the idea from a demo.
This keeps the work practical. It also gives search engines and AI answer engines a clean factual structure: what the topic is, who it helps, what to do next, and which risks matter before implementation.
Decision criteria for 1M context models tested
| Decision area | What to check | Production signal |
|---|---|---|
| Intent | Does 1M context models tested solve a real workflow problem? | The use case has a named owner and measurable outcome. |
| Data | Can the required data be used safely? | Sensitive data is classified and access is controlled. |
| Quality | Can a reviewer judge the output consistently? | Examples, rubrics, or acceptance criteria exist. |
| Scale | Can the workflow be repeated without hero effort? | The process is documented and can be handed to another team member. |
Practical example for 1M context models tested
A small business could use this article to choose one practical test. For example, a manager might take one customer-facing process, one internal document workflow, or one recurring content task and redesign only that step with AI support. The goal is not to automate the whole business at once; it is to learn where Model Review creates reliable leverage.
The useful deliverable is a short operating note: the trigger, the source material, the prompt or tool, the review checklist, the escalation rule, and the metric. That note becomes the handover asset for staff training, SEO/GEO content, service delivery, or future agent work.
Risks and controls for 1M context models tested
The common failure pattern is moving too quickly from a promising idea into an unmanaged workflow. For 1M context models tested, the risk is not only bad output. It can also be unclear data permission, staff confusion, duplicate content, unreviewed customer advice, or a tool that quietly changes cost or capability.
- Control tool sprawl with a named owner, a review step, and written acceptance criteria.
- Control unclear pricing with a named owner, a review step, and written acceptance criteria.
- Control vendor lock-in with a named owner, a review step, and written acceptance criteria.
- Control unreviewed data sharing with a named owner, a review step, and written acceptance criteria.
Measurement plan for 1M context models tested
A useful AI or SEO initiative should leave evidence. Track time to value, adoption rate, cost per workflow, quality review score and compare the pilot against the current process. If the measure does not improve, keep the learning but avoid scaling the workflow.
For GEO readiness, the page should also answer the core question directly, define the entities involved, include implementation steps, explain tradeoffs, and link readers to the next relevant AI Kick Start service, guide, tool, or article.
Definitions and entities for 1M context models tested
For search, GEO, and staff handover, define the core entities in plain language. In this article the important entities are the workflow owner, the AI tool or model, the source material, the review process, the risk boundary, and the measurable business outcome. Clear definitions make the page easier for people to scan and easier for AI answer engines to quote accurately.
- Workflow owner: the person accountable for deciding whether 1M context models tested belongs in the business process.
- Source material: the documents, examples, policies, URLs, prompts, videos, or customer questions that ground the output.
- Review boundary: the point where a human checks accuracy, privacy, brand voice, or customer impact before the result is used.
- Success metric: the measure that proves whether the tool evaluation workflow is worth repeating.
1M context models tested versus doing nothing
Doing nothing is also a decision. The cost may be slow manual work, weaker search visibility, inconsistent advice, duplicated effort, or staff using unmanaged AI tools without a shared process. The practical question is whether a controlled pilot can reduce that cost without creating a larger governance problem.
| Option | When it makes sense | What to watch |
|---|---|---|
| Do nothing | The workflow is rare, low value, or already reliable. | Competitors may improve speed, content depth, or service consistency first. |
| Run a small pilot | The task repeats often and has clear review criteria. | Keep scope tight and measure the result against the current process. |
| Build a production workflow | The pilot is repeatable and risk controls are documented. | Assign ownership, monitoring, training, and a rollback path. |
AI Kick Start handover package for 1M context models tested
A production handover should be concrete enough that another person can run it. For 1M context models tested, that means a short brief, a workflow map, approved prompts or tool settings, source material, a review checklist, internal links to supporting resources, and a simple measurement sheet. This is the difference between reading about AI and turning it into operational capability.
That packaging also strengthens E-E-A-T. It shows experience through implementation notes, expertise through decision criteria, authoritativeness through source-aware structure, and trust through risks, controls, and review steps. The article becomes useful even if the reader never buys a tool because it helps them make a better operational decision.
