Gemini 3.8 Flash outscores Claude Sonnet 5, runs four times faster and costs less, even after January's price rise.
Gemini 3.8 Flash scores 41 to Sonnet 5's 38 at 332 tokens per second against 79.5. At $1.50 / $7.50 from January it still undercuts Sonnet 5's $2 / $10.
GTM Architect & Growth Operator · Now · 16 September 2026
TL;DR · Key insights
- On Artificial Analysis's Intelligence Index v4.3, Gemini 3.8 Flash at high reasoning scores 41.19 and Claude Sonnet 5 at max effort scores 38. Gemini is ahead by about three points.
- Gemini 3.8 Flash generates 332 output tokens per second. Sonnet 5 generates 79.5. That is more than four times as fast.
- Today Gemini lists at $0.75 / $3.75 against Sonnet 5's $2 / $10. From 1 January 2027 Google lists $1.50 / $7.50, which is still 25% below Sonnet 5 on both sides.
- On paper Gemini 3.8 Flash wins every row. What keeps Sonnet 5 in the picture is the harness and the ecosystem around it, not the model's numbers.
In most comparisons Claude Sonnet 5 has at least one row to point to: against GPT-5.6 Terra it is four index points behind and $2 cheaper on output.
Against Gemini 3.8 Flash, none. The best Sonnet 5 manages in the table below is a tie on context window.
| Info | Gemini 3.8 Flash | Claude Sonnet 5 |
|---|---|---|
| Input / output per 1M, now | $0.75 / $3.75 | $2 / $10 |
| Input / output from 1 Jan 2027 | $1.50 / $7.50 | $2 / $10 |
| Intelligence Index v4.3 | 41.19 (high) | 38 (max) |
| Output speed, tokens/s | 332.0 | 79.5 |
| Blended price per 1M | $0.58 | $1.54 |
| Context window | 1M | 1M |
Prices from Google's Gemini API pricing page and Anthropic's pricing page. Index from Artificial Analysis's v4.3 announcement and Sonnet 5 model page; speed and blended price from its model pages. Read on 16 September 2026.
More than four times as fast
Gemini 3.8 Flash generates 332 output tokens per second on Artificial Analysis’s measurement, and Sonnet 5 generates 79.5.
Put in time, a 2,000-token answer takes about six seconds on Gemini. On Sonnet 5 it takes about 25. In Claude Code, or in any loop where someone waits for each response, that gap lands on every turn.
Anthropic’s own models table labels Sonnet 5 fast. Inside Anthropic’s lineup it is. Next to Gemini’s Flash tier it belongs to a slower class.
A few points ahead on the index
On Intelligence Index v4.3, Gemini 3.8 Flash at high reasoning scores 41.19 and Sonnet 5 at max effort scores 38.
Three points is a small lead. What it rules out is the usual trade, where a model this fast has given up capability to get there.
Cheaper now, and still cheaper in January
Today Gemini lists at $0.75 / $3.75 against Sonnet 5’s $2 / $10, about 2.7 times cheaper on both sides.
Google’s pricing page schedules a rise to $1.50 / $7.50 from 1 January 2027, and in Gemini 3.8 Flash against Luna I argued that any product should be priced on that number. Against Sonnet 5 it still wins. That is $1.50 against $2 on input and $7.50 against $10 on output: 25% cheaper on both.
Sonnet 5 had a rise scheduled too. Anthropic planned to move it to $3 / $15 on 1 September 2026 and then cancelled the increase, so $2 / $10 is now its standard price.
Price Gemini 3.8 Flash at its 2027 rate and it still comes out ahead on every number here. If Sonnet 5 is the right choice for you, the reason lies somewhere other than the model.
What Sonnet 5 still has
What it has is everything around the model, and for a team already on Anthropic that can outweigh the whole table:
- Claude Code. If your team already works in Claude Code, Sonnet 5 runs inside a harness whose permissions and conventions you have already set up.
- Prompts and tools tuned for Claude. Moving to another model family means re-testing prompts and tool definitions against a model that reads them differently. On a mature product that is weeks of work.
- Agreements. If your procurement or data-residency terms cover Anthropic and not Google, that settles it before any benchmark does.
- One vendor for escalation. A router that steps up from Sonnet 5 to Opus 5 or Fable 5.1 stays inside one API.
Those are real switching costs and a fair reason to stay put. What they measure is the price of moving. None of them says anything about which model is better.
Which one I would use
| Situation | Model | Why |
|---|---|---|
| New work, vendor is open | Gemini 3.8 Flash | Higher score, over four times faster, cheaper even at 2027 prices. |
| Latency-critical interactive product | Gemini 3.8 Flash | 332 against 79.5 tokens per second is felt on every response. |
| Team already working in Claude Code | Claude Sonnet 5 | The harness is the product you use; switching costs outweigh the gap. |
| Routing up to Opus 5 or Fable 5.1 | Claude Sonnet 5 | Keeping the escalation path in one API has real operational value. |
| Agreements cover Anthropic only | Claude Sonnet 5 | Vendor terms decide before benchmarks do. |
On the model alone, Gemini 3.8 Flash. On the environment around it, often Sonnet 5.
For Sonnet 5 against its nearest Claude alternatives, see Sonnet 5 against Opus 5 and Haiku 4.5 against Sonnet 5.
What would change my mind
A Sonnet 5 point release. The index gap is three points; speed is the number that would have to move.
Google raising the January price beyond what it lists. Gemini would need to reach $2 / $10 to lose its price lead, a third above the listed 2027 rate.
A like-for-like harness comparison. Most people meet Sonnet 5 inside Claude Code, and the index measures the model on its own.
Questions people asked
Is Gemini 3.8 Flash better than Claude Sonnet 5?
On published benchmarks, modestly. It scores 41.19 against 38 on the Intelligence Index v4.3, runs more than four times as fast, and is cheaper both now and after January.
Which is cheaper, Gemini 3.8 Flash or Claude Sonnet 5?
Gemini 3.8 Flash, at $0.75 / $3.75 now and $1.50 / $7.50 from 1 January 2027, against Sonnet 5’s $2 / $10. Even after the rise it is 25% cheaper on both sides.
Why would anyone choose Claude Sonnet 5 over Gemini 3.8 Flash?
Because of what surrounds the model. A team that works in Claude Code, with prompts tuned for Claude and a router that escalates to Opus 5, pays a real cost to move. So does one whose agreements cover Anthropic only.
Should I use Gemini 3.8 Flash or Claude Sonnet 5?
Gemini 3.8 Flash for new work where you can choose the vendor, and Sonnet 5 where your tooling or agreements are built around Anthropic.
If you are deciding whether a vendor switch pays for itself on your workload, the contact page is the fastest route to me.
