The Ultimate Gemini Model Comparison: 2.5 to 3.8 Flash, Pro & Flash-Lite

Google now runs two Gemini timelines that no longer move together. The Flash line has raced all the way to Gemini 3.8 Flash, launched September 2, 2026, while the Pro line has been frozen at Gemini 3.1 Pro since February. If you have been hunting for a “Gemini 3.5 Pro” or a “3.6 Pro,” that is the first thing to clear up in any honest Gemini model comparison, because neither one exists yet.

This guide maps the entire lineup from Gemini 2.5 through 3.8 Flash, one version at a time, with a benchmark and pricing table for each, plus how the models surface inside the Gemini app and a plain recommendation for every use case. You will see exactly why Google split the family in two, what actually changed at each release, and which model is the right default for coding, everyday chat, long documents, and tight budgets.

The Key Takeaways

• Gemini 3.8 Flash is the newest model, live since September 2, 2026, at an introductory $0.75 / $3.75 per 1M tokens. It is Google’s third Flash release in 43 days, and 3.7 Flash and 3.6 Flash sit on the identical rate.
• There is no Gemini 3.5 Pro or 3.6 Pro. The current flagship Pro is still Gemini 3.1 Pro from February 2026.
• That cheap rate expires. All three of 3.8, 3.7 and 3.6 Flash go back to $1.50 / $7.50 on January 1, 2027, so budget past the new year on the higher number.
• Gemini 3.5 Flash-Lite is the cheapest current model at $0.30 / $2.50 per 1M tokens.
• In the app, the Free plan runs Gemini 3.6 Flash. Neither 3.7 nor 3.8 Flash is free: 3.7 reaches consumers only through Spark, and 3.8 access so far is reported for AI Pro and Ultra subscribers. Deep Think stays on AI Ultra, while Gemini Spark reaches AI Pro in the US and 160+ more countries.
• Google confirmed it has begun its “most ambitious pre-training run yet” for Gemini 4.

The Gemini lineup at a glance (2026)

Google organizes Gemini into three tiers. Pro is the flagship reasoning tier, Flash is the fast everyday workhorse, and Flash-Lite is the budget option. The confusing part is that these tiers no longer share a version number, so the newest Flash (3.8) is four full releases ahead of the newest Pro (3.1).

Here is the current family, oldest to newest, with the models that matter most in bold.

Two specialist models sit outside the main grid. Gemini 3 Deep Think is an extended-reasoning preview from December 2025, and Gemini 3.5 Flash Cyber, released July 21, is a version of 3.5 Flash fine-tuned to find and patch security vulnerabilities inside Google’s CodeMender system.

Why there’s no Gemini 3.5 Pro (or 3.6 Pro)

The single biggest source of confusion in any Gemini model comparison is the missing Pro releases. When Google shipped 3.5 Flash, then 3.6 Flash, then 3.7 Flash, then 3.8 Flash, many people assumed a matching Pro arrived too. It did not.

According to TechCrunch, Google released three new models on July 21 and pointedly no 3.5 Pro. Bloomberg reporting cited in that coverage says Google hit internal delays and struggled to meet its own performance goals for 3.5 Pro, so it remains in partner testing rather than general release. Product lead Logan Kilpatrick said the company is still testing it and hopes it will “land soon.”

That leaves Gemini 3.1 Pro, from February 2026, as the newest Pro model you can actually use. Meanwhile the Flash line kept shipping, which is why the version numbers look mismatched. The takeaway is simple, when you want top-tier Gemini reasoning today, you use 3.1 Pro, not a 3.5 or 3.6 Pro that has not been released.

Gemini 2.5: the legacy three-tier foundation

The 2.5 family is the legacy generation, released across 2025, and it set the three-tier pattern Google still uses. Gemini 2.5 Pro arrived March 25, 2.5 Flash on April 17, and 2.5 Flash-Lite on June 17, each with a 1M-token context window. These models still work and remain a cheaper fallback, but they trail the 3.x family on reasoning, coding, and agentic tasks.

How the 2.5 tiers benchmark and price

At launch, 2.5 Pro topped the LMArena leaderboard and scored 63.8% on SWE-Bench Verified. 2.5 Flash undercut it sharply, and 2.5 Flash-Lite went lower still at $0.10 / $0.40, still the cheapest Gemini rate the API has ever carried.

When to pick a 2.5 model

Use 2.5 only if you are on older infrastructure or need the lowest possible cost and do not care about frontier quality. For anything new, the 3.x tiers are a clear upgrade at similar or lower prices.

Gemini 3.0: the generation reset

Model Gemini 3 opened the current generation. Gemini 3 Pro launched November 18, 2025 as a preview flagship, followed by Gemini 3 Deep Think on December 4 for harder multi-step reasoning, and Gemini 3 Flash on December 17. This was the jump that reset Google’s benchmark position against ChatGPT and Claude.

Gemini 3.0 benchmarks and pricing

Gemini 3 Pro shipped with a 1M-token context window and 64K output, posting 91.9% on GPQA Diamond, a 1501 Elo on LMArena, and 76.2% on SWE-Bench Verified. Gemini 3 Flash then beat its own Pro sibling on coding, hitting 78% on SWE-Bench Verified at just $0.50 / $3.00 per 1M tokens. Deep Think, an extended-reasoning mode for AI Ultra subscribers, pushed ARC-AGI-2 to 45.1% with code execution.

How the 3.0 generation holds up

The 3.0 models introduced the architecture and long-context work that later versions built on. Most have since been superseded by 3.1 and the newer Flash releases, so treat 3.0 as the foundation rather than a model you would pick today.

Gemini 3.1 Pro: the current flagship

Gemini 3.1 Pro, released February 19, 2026, is the model to reach for when you need maximum capability. It uses a sparse mixture-of-experts design, handles a 1M-token context window, and outputs up to 64K tokens in a single response. It scores 94.3% on GPQA Diamond and 80.6% on SWE-Bench Verified, though on the Artificial Analysis Intelligence Index v4.1 it lands at 46, behind the current OpenAI and Anthropic flagships.

Gemini 3.1 Pro specs and pricing

When to use 3.1 Pro

It is the right choice for complex reasoning, long-document analysis, research, and the hardest agentic coding tasks. Because Google has not updated the Pro tier since February, 3.1 Pro is likely to remain the flagship until 3.5 Pro finally clears testing.

Gemini 3.5 Flash and Flash-Lite: the mid-2026 workhorses

Gemini 3.5 Flash shipped at Google I/O on May 19, 2026, at $1.50 / $9.00 per 1M tokens with a 1M-token context window, and it notably beat 3.1 Pro on several coding and agentic benchmarks despite being a Flash model. It served as the everyday workhorse until 3.6 replaced it in July.

Gemini 3.5 Flash-Lite: the budget champion

Gemini 3.5 Flash-Lite, released July 21 at $0.30 / $2.50 per 1M tokens, is the current budget champion. Google says it beats the older, larger Gemini 3 Flash outright on some evals, including SWE-Bench Pro and OSWorld-Verified, which makes it a strong price-to-performance pick for high-volume work.

Why the Flash pairing matters

The pairing is the point. Flash covers quality everyday work while Flash-Lite handles the high-volume, cost-sensitive jobs, and both hold a full million-token window. When 3.6 Flash arrived in July it slotted in above 3.5 Flash, 3.7 Flash took the top of the Flash line three weeks later, and 3.8 Flash took it three weeks after that, while Flash-Lite stayed as the value floor throughout.

Gemini 3.8 Flash: the current workhorse

Gemini 3.8 Flash went generally available on September 2, 2026, three weeks after 3.7 Flash and Google’s third Flash release in 43 days. Every line of the spec sheet carries over unchanged from its predecessor: the same 1M-token context window, the same 64K output limit, the same March 2026 knowledge cutoff, the same modality list and the same launch price. The entire release is about what the model does inside that envelope, and Google tuned this one for long-horizon agents rather than the raw coding focus of 3.7.

Gemini 3.8 Flash pricing and benchmark gains

The price did not move. 3.8 Flash lists at the same introductory $0.75 / $3.75 per 1M tokens as 3.7 and 3.6 Flash, and it carries the same footnote: on January 1, 2027 all three go to $1.50 / $7.50. It beats 3.7 Flash on every row Google published, with the biggest gains on BioMysteryBench Human Difficult (43.5% to 56.5%), Terminal-bench 4.0 (11.2% to 19.1%) and DeepSWE v1.1 (65.3% to 73.7%). It does not sweep the field, though: Claude Opus 5 still wins the hardest agentic tests, taking Terminal-bench 4.0 51.8% to 19.1% and OSWorld-2.0 75.4% to 59.0%. Our Gemini 3.8 Flash breakdown has the full 14-row table and the widely republished benchmark figure that turned out to be wrong.

Gemini 3.7 Flash: the previous workhorse

Gemini 3.7 Flash arrived on August 13, 2026 — just 23 days after 3.6 Flash — in AI Studio, the Gemini API, Android Studio and the Gemini Enterprise Agent Platform, with the same 1M-token context window and 64K output limit as its predecessor. Unlike the 3.6 release, this one was a genuine capability jump rather than an efficiency pass: coding and agentic scores moved a long way in three weeks. It held the top of the Flash line for exactly three weeks before 3.8 Flash replaced it at the same price.

Gemini 3.7 Flash pricing and benchmark gains

The price halved to $0.75 / $3.75 per 1M tokens, and Google put 3.6 Flash on the identical rate, which is also where 3.8 Flash later landed, so all three cost the same today. Read the footnote before budgeting: those rates are introductory and expire on December 31, 2026, rising to $1.50 / $7.50 in January. On benchmarks, DeepSWE v1.1 climbed from 49.0% to 65.3% and AutomationBench nearly doubled from 17.0% to 30.4%. For the full picture, including the regions Google locked out, see our Gemini 3.7 Flash breakdown.

Gemini 3.6 Flash: the previous workhorse

Gemini 3.6 Flash shipped on July 21, 2026 and remains the default model in the free Gemini app, with a 1M-token context window. It is an efficiency upgrade rather than a new frontier tier. It uses about 17% fewer output tokens than 3.5 Flash on the Artificial Analysis Index, with reductions up to 65% on some long-horizon engineering tasks.

Gemini 3.6 Flash pricing and benchmark gains

The price change at launch was often misreported. The cut was output-only: output dropped from $9.00 to $7.50 per 1M tokens while input stayed flat at $1.50. Google has since cut the model again, to $0.75 / $3.75, to match Gemini 3.7 Flash. On benchmarks, DeepSWE rose from 37% to 49%, MLE-Bench from 49.7% to 63.9%, and OSWorld-Verified from 78.4% to 83.0%, per Google’s announcement. For a deeper look at this release, see our full Gemini 3.6 Flash breakdown.

Gemini API pricing compared (2026)

Pricing is where the tier split becomes practical. Flash and Flash-Lite are built for volume, while Pro costs more and is reserved for the tasks that justify it. The table below shows current API rates per 1 million tokens.

For most workloads, Gemini 3.8 Flash now delivers the best results in the Flash line, and because 3.6 and 3.7 Flash sit on the same rate there is no price argument for staying on an older model. All three Flash rates are introductory and double on January 1, 2027. If you are running huge volumes of simpler requests, 3.5 Flash-Lite cuts costs further while holding surprisingly strong quality.

Note that the Pro tier is metered by prompt size. The $2.00 / $12.00 rate covers prompts up to 200K tokens and doubles to $4.00 / $18.00 above that, which is why the flagship is reserved for work that genuinely needs it.

How Gemini appears in the app (Free, Plus, Pro, Ultra)

Most people never touch the API. Inside the Gemini app the models show up as consumer plans rather than version numbers, and which release you get depends on what you pay. The Free plan still runs Gemini 3.6 Flash by default with only limited access to the Pro reasoning model, while the paid tiers raise your usage limits and open up the heavier features. Note that neither newer Flash release appears as a plan tier: 3.7 Flash reaches app users only inside Spark, which needs AI Pro or Ultra, and 3.8 Flash has been reported rolling out to AI Pro and Ultra subscribers even though the Gemini Apps release notes carry no entry for it yet.

The practical read is simple. If you just want a fast, capable assistant, the Free plan and its 3.6 Flash engine cover most needs. Step up to AI Pro for heavier reasoning, Deep Research and the Gemini Spark agent, which reached Pro in the US on July 23, 2026 and more than 160 additional countries on July 30. Only the AI Ultra plans add Deep Think. Spark is not offered at all in the EEA, the UK, Switzerland or Nigeria.

Gemini benchmarks compared

Benchmarks tell the same split-lineage story. On the Artificial Analysis Intelligence Index v4.1, a composite of nine evaluations, the Flash models now edge out the older Pro. Both Gemini 3.5 Flash and Gemini 3.6 Flash score 50, while the flagship Gemini 3.1 Pro sits at 46 and the budget Gemini 3.5 Flash-Lite lands at 36.

One note on reading that table. Artificial Analysis has since moved to a newer revision of the index, v4.1.1, on which Gemini 3.8 Flash scores 59 at the high reasoning setting, ranking 16th of 195 models, against 56 for 3.7 Flash and 52 for 3.6 Flash. Those are not comparable to the v4.1 figures below, which is why they sit in separate rows rather than replacing them. Artificial Analysis scores are rolling medians that move week to week, so always check the index version, the reasoning setting and the date attached to any figure you see quoted, including the older ones on this page.

The flat 50 between 3.5 and 3.6 Flash is the key detail. Gemini 3.6’s real gains show up in efficiency and agentic coding rather than the composite score. As noted above, it posts double-digit jumps on DeepSWE and MLE-Bench while using about 17% fewer output tokens than 3.5 Flash. For raw speed, Flash-Lite is in another class at 463 tokens per second, versus 189 for 3.5 Flash and 113 for 3.1 Pro.

Benchmark progression across the Gemini family

Tracking a few key evals across releases shows the two tracks clearly. The Pro line climbs steadily on the hardest reasoning and coding tests, while the Flash line quietly catches the flagship on the composite index. Empty cells mean Google or Artificial Analysis has not published that number for the model.

Read down the Pro columns and the trend is a clean climb, GPQA Diamond from 91.9% to 94.3% and SWE-Bench Verified from 63.8% through 76.2% to 80.6%. Read across the bottom rows and you see why Flash gets the attention, it matches the flagship’s composite score of 46 to 50 while its agentic coding jumps sharply from 37% to 49% on DeepSWE, and 3.8 Flash carries that to 73.7% on the v1.1 revision of the same test. The Pro tier still owns peak reasoning, but Flash has closed most of the everyday gap.

Which Gemini model should you use?

For most people, Gemini 3.8 Flash is the right default. It is fast, cheap at $0.75 / $3.75 per 1M tokens until January, and handles coding and everyday tasks well. Step up to Gemini 3.1 Pro for the hardest reasoning and long-document work, and drop to Gemini 3.5 Flash-Lite when cost matters most. The table below maps common jobs to the model that fits.

If your main job is programming, 3.8 Flash is the value pick and 3.1 Pro is the ceiling for the toughest problems. One caveat for non-developers: neither 3.7 nor 3.8 Flash is free in the Gemini app, so the free tier still answers you with 3.6 Flash. For a broader look at matching tasks to models across vendors, our guide on when to use which AI model covers the full field.

How Gemini stacks up against ChatGPT and Claude

Google’s strategy differs from its rivals. Instead of chasing a single maximum-benchmark flagship, it optimizes for practical versatility, multimodality, and deep integration across Search, Workspace, and Android. That is why the Flash tier gets so much attention, it is the model most people actually touch every day.

The gap is real but narrower than the raw scores suggest. Gemini 3.1 Pro trails GPT-5.6 Sol at 59 and Claude Opus 5 at 61 on the composite v4.1 index, yet it is also the cheapest of the three by a wide margin, roughly half the input price and well under half the output price. Google’s bet is that most work does not need the very top of the reasoning curve, and its Flash tier, which scored 50 for 3.6 Flash on that same v4.1 index, closes most of that gap for everyday tasks at a fraction of the cost.

All three flagships now handle around a million tokens of context, so the real trade-off is reasoning depth and price rather than window size. One practical difference, OpenAI and Anthropic allow far longer single responses at 128K output tokens versus Gemini’s 64K, which matters for large code generation or long reports.

On raw frontier reasoning, 3.1 Pro competes with the top GPT-5.x and Claude models without leading the pack, while Flash punches above its price class. For the full cross-vendor picture, see our best AI models comparison and the deep-dive Gemini 3.5 review.

What’s next for Gemini (3.5 Pro and Gemini 4)

Two releases hang over the current lineup. The first is Gemini 3.5 Pro, still stuck in partner testing after Google missed its own internal goals, with the product team saying only that it hopes the model will “land soon.” When it ships, it will finally give the Pro tier its overdue update and likely reset the top of this comparison.

The second is Gemini 4. Alongside the July 2026 releases, Google confirmed it has begun its “most ambitious pre-training run yet,” a strong signal that the next full generation is in active development. No date has been announced, and 3.5 Pro is still expected to arrive first, so the two-track map is likely to persist for a while yet.

Use every top Gemini-class model without picking a version

The version maze this article just mapped is the real cost of picking a Gemini tier. Google AI Pro is $19.99 a month and still answers you with whichever release the app decides you get, and on the API side you are tracking which Flash rate is introductory and which Pro threshold doubles your bill.

Fello AI is a native Mac, iPhone and iPad app that gives you Gemini behind one subscription, along with ChatGPT, Claude, Grok, DeepSeek, Perplexity, Kimi, GLM and Qwen. It starts at $9.99 a month, half of Google AI Pro, with a free tier to try first and a 4.7-star rating across 27,000+ reviews. The current frontier Gemini model is simply there when you open it, and you can put the same question to a rival model in the same conversation instead of committing to a version and a price table.

Conclusion

The honest Gemini model comparison for 2026 is a story of two tracks. Flash has sprinted to 3.8, three releases in 43 days and cheaper than ever, while Pro sits patiently at 3.1 waiting for 3.5 to clear testing. For nearly everyone, start with Gemini 3.8 Flash, step up to 3.1 Pro for the hardest work, and drop to 3.5 Flash-Lite when budget rules. With Gemini 4 already in pre-training, expect the map to shift again before long.

FAQ