Gemini 3.6 Flash launched on July 21, 2026, and it is a direct upgrade to Gemini 3.5 Flash. Same API, same context window, same input modalities. But cheaper on output tokens, faster, and better at coding and agentic tasks.
The short answer: yes, you should switch. The API is identical, just change the model name from gemini-3.5-flash to gemini-3.6-flash. There is no reason to stay on 3.5 Flash unless you have specific compatibility constraints.
Here is the full breakdown of what changed and what it means for your code.
Spec Comparison
| Spec | 3.6 Flash | 3.5 Flash |
|---|---|---|
| Release date | July 21, 2026 | May 19, 2026 |
| Context window | 1M tokens | 1M tokens |
| Max output | 65,536 tokens | 65,536 tokens |
| Input modalities | Text, Image, Video, Audio, PDF | Text, Image, Video, Audio, PDF |
| Output | Text only | Text only |
| Knowledge cutoff | March 2026 | January 2025 |
| Speed | 304 tok/s | 289 tok/s |
| Input price | $1.50/1M | $1.50/1M |
| Output price | $7.50/1M | $9.00/1M |
| Cached input | $0.15/1M | $0.15/1M |
| Computer use | Built-in | Not supported |
| Thinking mode | Yes | Yes |
The specs are nearly identical. The differences are: cheaper output, faster speed, newer knowledge cutoff, and computer use.
Pricing: 3.6 Flash Is Cheaper
The input price is the same: $1.50/1M. But output dropped from $9.00 to $7.50 per 1M tokens. That is a 17% reduction on output cost.
For an agent workflow that generates 10,000 output tokens per task:
- 3.5 Flash: $0.09 per task
- 3.6 Flash: $0.075 per task
That does not sound like much, but at scale it adds up. If you run 100,000 tasks per month, you save $1,500/month on output tokens alone.
Google also claims 3.6 Flash uses 17% fewer output tokens to accomplish the same tasks. If that holds, the effective cost per task drops even further:
- 3.5 Flash: 10,000 tokens x $9.00/1M = $0.090
- 3.6 Flash: 8,300 tokens x $7.50/1M = $0.062
That is a 31% reduction in cost per task. Real-world numbers will vary by workload, but the direction is clear.
Benchmarks: Where 3.6 Flash Wins
| Benchmark | 3.6 Flash | 3.5 Flash | Change |
|---|---|---|---|
| DeepSWE (Datacurve) | 49% | 37% | +12 points |
| MLE Bench | 63.9% | 49.7% | +14.2 points |
| OSWorld-Verified | 83.0% | 78.4% | +4.6 points |
| GDPval-AA v2 | 1421 | 1349 | +72 Elo |
The coding benchmarks show the biggest improvements. DeepSWE measures real-world software engineering, and a 12-point jump means fewer unwanted code edits and fewer execution loops. MLE Bench tests ML research tasks, where 3.6 Flash is significantly stronger.
The OSWorld-Verified improvement (computer use) is notable because 3.5 Flash did not have built-in computer use at all. This is a new capability, not just an incremental improvement.
Speed: Marginal But Meaningful
304 tok/s vs 289 tok/s is a 5% improvement. Not a huge jump, but 3.6 Flash was already the fastest frontier-tier model. At 304 tok/s, it is roughly 2x faster than Claude Sonnet 5 (~180 tok/s) and 4x faster than GPT-5.6 Sol (~150 tok/s).
The bigger speed win comes from token efficiency. If 3.6 Flash uses 17% fewer tokens per task, the wall-clock time per task drops even if the raw token rate only improved by 5%.
Computer Use: The New Feature
This is the biggest feature addition. 3.5 Flash had no computer use capability. 3.6 Flash ships with built-in computer use via the Gemini API.
The model can now interact with browser and desktop environments through screenshots and mouse/keyboard actions. Google’s benchmark shows 83.0% on OSWorld-Verified, which tests the model’s ability to complete tasks in a computer environment.
For agent builders, this opens up workflows that were impossible with 3.5 Flash: filling out forms, navigating web apps, interacting with legacy systems that only have a GUI.
Knowledge Cutoff: March 2026
3.5 Flash had a January 2025 cutoff. 3.6 Flash advances this to March 2026. The model now knows about events through early 2026 without needing Search as a tool enabled.
This matters for tasks that reference recent libraries, frameworks, or APIs. If you were hitting the knowledge cutoff on 3.5 Flash, 3.6 Flash fixes that.
What Has Not Changed
- Context window: 1M tokens (same)
- Max output: 65,536 tokens (same)
- Input modalities: Text, Image, Video, Audio, PDF (same)
- Output: Text only (same)
- Thinking mode: Yes (same)
- Cached input: $0.15/1M (same)
- Function calling, structured output, Search as a tool, Code execution: All the same
The API is identical. If your code works with gemini-3.5-flash, it will work with gemini-3.6-flash just by changing the model name.
How to Switch
# Before
response = client.models.generate_content(
model="gemini-3.5-flash",
contents="Your prompt here"
)
# After
response = client.models.generate_content(
model="gemini-3.6-flash",
contents="Your prompt here"
)That is it. No code changes needed beyond the model name. The API, SDK, and response format are all the same.
When to Stay on 3.5 Flash
There are very few reasons to stay on 3.5 Flash:
- Regression testing. If you have tests that depend on specific 3.5 Flash behavior, switching might break them. Run your test suite on 3.6 Flash first.
- Benchmark reproducibility. If you published benchmark results using 3.5 Flash and need to keep them consistent.
- Rate limits. 3.6 Flash might have different rate limits during the initial rollout period.
For everyone else, switch now.
Verdict
| Category | Winner |
|---|---|
| Price | 3.6 Flash |
| Speed | 3.6 Flash |
| Coding | 3.6 Flash |
| Computer use | 3.6 Flash |
| Knowledge cutoff | 3.6 Flash |
| Token efficiency | 3.6 Flash |
| API compatibility | Tie |
3.6 Flash wins on every dimension that matters. The switch is a no-brainer.
My Take
This is the easiest upgrade decision in the Gemini lineup. There is literally no reason to stay on 3.5 Flash unless you have regression tests that depend on specific 3.5 Flash behavior. The API is identical, the price is lower, the speed is higher, and the coding benchmarks are better.
If you are building agents, the built-in computer use alone is worth the switch. If you are doing high-volume work, the 17% token efficiency improvement means real cost savings.
The only thing to watch: if you are comparing Gemini to other models, do not use 3.5 Flash benchmarks as your reference point. Use 3.6 Flash. The landscape has shifted. See Gemini 3.6 Flash vs Claude Sonnet 5 and Gemini 3.6 Flash vs GPT-5.6 for how it stacks up against the competition.
For the full specs, see the Gemini 3.6 Flash complete guide. For pricing context, see AI API Pricing Compared 2026.
FAQ
Is 3.6 Flash a free upgrade from 3.5 Flash?
Yes. Same pricing tiers, same subscription plans. You get better performance at a lower output cost.
Will 3.5 Flash be deprecated?
Google has not announced a deprecation date for 3.5 Flash. It will likely remain available for some time. But there is no reason to use it now that 3.6 Flash exists.
Does 3.6 Flash support the same tools as 3.5 Flash?
Yes. Function calling, structured output, Search as a tool, Code execution, and now computer use. Everything 3.5 Flash supports, 3.6 Flash supports too.
Is the thinking mode different?
No. Same thinking mode, same thinking_config parameter, same behavior. 3.6 Flash may produce different thinking content because it is a different model, but the API is identical.
Can I use 3.6 Flash in Gemini CLI?
Yes. Update your Gemini CLI to the latest version and it should default to 3.6 Flash. See our Gemini CLI guide for details.