Gemini 3.8 Flash and Muse Spark 1.3 Pricing
Gemini 3.8 Flash costs $0.75 per million input tokens until December 31, then doubles. Muse Spark 1.3 is $1.25, or $0.10 if Meta can train on you.
Two models shipped on September 2, 2026, and neither one changed a price. Google released Gemini 3.8 Flash and Gemini 3.8 Flash Cyber at the rate Gemini 3.7 Flash already charged. Meta released Muse Spark 1.3 at the rate Muse Spark 1.2 already charged.
The price that moves is the one already on the calendar. Google’s Flash rate is introductory, and the Gemini API pricing page states it doubles on January 1, 2027. Budget the coming year on the number in today’s launch coverage and the figure will be wrong by two thirds.
How much does Gemini 3.8 Flash cost?
Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens, with cached input at $0.075. Google’s own pricing page marks all three as introductory rates that run “through December 31, 2026” and rise on “January 1, 2027” to $1.50, $7.50 and $0.15.
| Line item | Through Dec 31, 2026 | From Jan 1, 2027 |
|---|---|---|
| Input, per Mtok | $0.75 | $1.50 |
| Output, per Mtok | $3.75 | $7.50 |
| Cached input read, per Mtok | $0.075 | $0.15 |
| Cache storage, per Mtok-hour | $0.50 | $1.00 |
Everything doubles. Not the headline rate with the cache rate held back, and not input with output held back: every line on the card, on one date.
The model itself is a 1M-token context window with a 64K maximum output, text, image, audio and video in, and a March 2026 knowledge cutoff, per the Gemini 3.8 Flash model card. Google reports 54.9% on HLE-Verified and says the model outperforms most larger frontier models on DeepSWE v1.1 for long-horizon software engineering. Those are vendor-reported figures from the launch announcement and have not been independently reproduced.
The Cyber variant is not a product most readers can buy. Google restricts Gemini 3.8 Flash Cyber to its Fairwind Program for trusted government authorities, critical infrastructure operators and software maintainers, and publishes no separate token rate for it.
How much does Muse Spark 1.3 cost?
Muse Spark 1.3 costs $1.25 per million input tokens and $4.25 per million output tokens on the standard Meta Model API endpoint, with cached input at $0.15. The Meta Model API pricing and rate limits page also lists a muse-spark-1.3-contributor endpoint at $0.10 input, $0.20 output and $0.002 cached input, in exchange for Meta using the prompts and completions sent through it.
| Line item | Standard | Contributor | Ratio |
|---|---|---|---|
| Input, per Mtok | $1.25 | $0.10 | 12.5x |
| Output, per Mtok | $4.25 | $0.20 | 21.25x |
| Cached input read, per Mtok | $0.15 | $0.002 | 75x |
That rate card is identical to the one Muse Spark 1.2 has carried since August, which this site covered in Muse Spark 1.2 pricing and cost per task. Meta moved capability and held price: Artificial Analysis, an independent benchmarking service, scores the generally available xhigh configuration at 61 on its Intelligence Index against 57 for 1.2 in the same configuration. A partner-only max configuration scores 62 and remains in limited preview pending further safety testing.
Meta also reports that 1.3 completes coding work with roughly 20% fewer tool calls and 25% fewer tokens than 1.2. If that holds on real workloads it is a price cut that never appears on a rate card, since the bill is tokens multiplied by rate and only one of those two terms is published.
The whole Flash tier expires on the same day
The interesting part of Google’s pricing is not that one model has an introductory rate. It is that three consecutive generations share it.
Gemini 3.6 Flash, Gemini 3.7 Flash and now Gemini 3.8 Flash all list $0.75 and $3.75 per million tokens, and all three carry the same December 31, 2026 expiry. The model registry on this site records the same trio of numbers three times over.
What a year of Gemini 3.8 Flash actually costs
Rates are not bills. To turn one into the other, take the multi-step agentic profile the cost-per-task calculator on this site already used before either model existed: 3 million input tokens, 90% of them cache hits, 80,000 output tokens, one pass. Run five of those a day across 22 working days.
At the introductory rate, one such task costs $0.73 on Gemini 3.8 Flash and the month costs $80. From January 1 the identical task costs $1.46 and the month costs $160. Muse Spark 1.3 sits between the two at $1.12 per task and $123 per month, and does not have a scheduled increase.
| Date | Gemini 3.8 Flash | Muse Spark 1.3 | Muse Spark contributor |
|---|---|---|---|
| 2026-09-01 | $80 | $123 | $6 |
| 2026-10-01 | $160 | $246 | $11 |
| 2026-11-01 | $240 | $370 | $17 |
| 2026-12-01 | $320 | $493 | $23 |
| 2027-01-01 | $480 | $616 | $28 |
| 2027-02-01 | $640 | $739 | $34 |
| 2027-03-01 | $800 | $862 | $40 |
| 2027-04-01 | $960 | $986 | $45 |
| 2027-05-01 | $1,120 | $1,109 | $51 |
| 2027-06-01 | $1,280 | $1,232 | $57 |
| 2027-07-01 | $1,440 | $1,355 | $62 |
| 2027-08-01 | $1,601 | $1,478 | $68 |
Two readings come out of that chart.
The first is a budgeting error waiting to happen. Twelve months of this workload priced at the introductory rate comes to $960. The actual twelve months, four at the old rate and eight at the new one, come to $1,601. Anyone sizing a 2027 line item from a September rate card is short by $641, or 67%.
The second is a ranking that inverts. Gemini 3.8 Flash is 35% cheaper per task than Muse Spark 1.3 today. From January it is 30% more expensive. Cumulative spend crosses in May 2027, so a nine-month commitment made this week on the basis of the cheaper number ends up on the more expensive model. Neither vendor’s launch material frames it that way, because neither vendor’s launch material mentions the other.
The comparison assumes Meta holds its rate, which is an assumption rather than a promise. It is a reasonable one: Muse Spark 1.1, 1.2 and 1.3 have all shipped at $1.25 and $4.25.
What the $0.10 tier actually costs
The contributor endpoint is the cheapest way to run a frontier-adjacent model, by a distance nothing else on this list approaches. The same agentic task costs $0.05 there against $1.12 on the standard endpoint, a factor of 21.8. A year of the workload above costs $68 instead of $1,478.
The price of that is stated plainly in Meta’s own documentation: prompts and completions sent to the contributor endpoint may be used to improve Meta’s products. For a hobby project or a public-data pipeline, that is close to free money. For anything touching a private codebase, customer records or unreleased work, it is not a pricing decision at all, and the discount is irrelevant because the option is unavailable.
| Dimension | Gemini 3.8 Flash | Muse Spark 1.3 | Muse Spark contributor |
|---|---|---|---|
| Cost per agentic task, today | $0.73 | $1.12 | $0.05 |
| Cost per agentic task, 2027 | $1.46, scheduled | $1.12, no change filed | $0.05, no change filed |
| Rate stability | Doubles Jan 1, 2027 | Flat over 3 releases | Flat over 2 releases |
| Your prompts and outputs | No training trade offered | No training trade offered | Used to improve Meta products |
| Independent index score | 59, high effort | 61, xhigh effort | 61, same model |
| Output speed | 306 tokens per second | 186 tokens per second | 186, same model |
Where the two land on independent scoring
Vendor benchmarks at launch are marketing artifacts. The useful cross-check is a third party running the same suite against everything.
Artificial Analysis, whose release pages were read on September 3, 2026, puts Muse Spark 1.3 at 61 on its Intelligence Index in the generally available xhigh configuration, tied with GPT-5.6 Sol at max effort and Grok 4.6 at high effort, and behind Claude Fable 5.1. It puts Gemini 3.8 Flash at 59 at high effort, up from 56 for Gemini 3.7 Flash. On the same service’s cost-to-run measure, the full index costs $0.58 to run on Gemini 3.8 Flash and $0.55 on Muse Spark 1.3, against $2.34 on Claude Opus 5.
Effort configuration matters more than the two-point gap between them. Artificial Analysis publishes several per model, and a score quoted without its configuration is not a comparable number. The 62 that some launch coverage attributes to Muse Spark 1.3 belongs to the max configuration, which Meta has not released to the public.
Speed splits them harder than intelligence does. Gemini 3.8 Flash runs at 306 output tokens per second against 186 for Muse Spark 1.3, which is the gap that shows up in an interactive agent loop rather than on a bill. Neither model is scored on the value leaderboard here yet, because the composite that board uses has not been published for either.
Which one to pick
For work that ships before the end of the year, Gemini 3.8 Flash is the cheaper model and the faster one, and the January date is somebody else’s problem. Take it.
For anything that runs past January 1, price the decision at $1.50 and $7.50, because that is what the vendor says the meter will read. On those numbers Muse Spark 1.3 is cheaper per task, and holds an index score two points higher.
For work with no confidentiality constraint at all, the contributor endpoint is not in the same price class as either. It is worth pricing the alternative honestly rather than dismissing it: $68 a year against $1,478 buys a lot of tolerance for a data-use clause, as long as nothing sensitive passes through it.
Both models are now in the model release archive with the rate each launched at, and in the September 2026 release roundup. To run the arithmetic above on a token mix that matches your own workload rather than the presets used here, the cost-per-task tool takes the same inputs.
Frequently asked questions
- How much does Gemini 3.8 Flash cost?
- Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens, with cached input at $0.075. Google states these are introductory rates through December 31, 2026, after which they double to $1.50 and $7.50, with cached input at $0.15.
- When does Gemini 3.8 Flash pricing go up?
- January 1, 2027. Every line on the card doubles on that date: input, output, cached input and hourly cache storage. The same expiry applies to Gemini 3.6 Flash and Gemini 3.7 Flash, which carry identical rates.
- How much does Muse Spark 1.3 cost?
- Muse Spark 1.3 costs $1.25 per million input tokens and $4.25 per million output tokens on the standard Meta Model API endpoint, with cached input at $0.15. That is unchanged from Muse Spark 1.2. A contributor endpoint costs $0.10 and $0.20 per million tokens in exchange for Meta using prompts and completions to improve its products.
- Is Gemini 3.8 Flash cheaper than Muse Spark 1.3?
- Today, yes: about 35% cheaper on a modeled multi-step agentic task, $0.73 against $1.12. From January 1, 2027, no: the same Gemini task costs $1.46, about 30% more than Muse Spark 1.3, assuming Meta holds its current rate.
- Which scores higher on independent benchmarks?
- Muse Spark 1.3, narrowly. Artificial Analysis scores its generally available xhigh configuration at 61 on the Intelligence Index against 59 for Gemini 3.8 Flash at high effort, read September 3, 2026. Gemini 3.8 Flash is substantially faster at 306 output tokens per second against 186.
- What is Gemini 3.8 Flash Cyber?
- A restricted variant tuned for vulnerability discovery and mitigation, scoring 47.2% pass@1 on CWE-Bench. Google limits access to its Fairwind Program for trusted government authorities, critical infrastructure operators and software maintainers, and publishes no separate token rate for it.
Sources
Google (2026). Introducing Gemini 3.8 Flash and 3.8 Flash Cyber. Google blog announcement. https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/ Verified 2026-09-03.
Google (2026). Gemini Developer API pricing. Google AI for Developers documentation. https://ai.google.dev/gemini-api/docs/pricing Verified 2026-09-03.
Google DeepMind (2026). Gemini 3.8 Flash model card. Google DeepMind documentation. https://deepmind.google/models/model-cards/gemini-3-8-flash/ Verified 2026-09-03.
Meta AI Research (2026). Introducing Muse Spark 1.3. Meta AI Research blog. https://research.meta.ai/blog/introducing-muse-spark-1-3 Verified 2026-09-03.
Meta (2026). Model API pricing and rate limits. Meta Model API documentation. https://dev.meta.ai/docs/pricing-rate-limits/ Verified 2026-09-03.
Artificial Analysis (2026). Muse Spark 1.3: Meta reaches the frontier. Artificial Analysis, independent benchmarking service. https://artificialanalysis.ai/articles/muse-spark-1-3 Verified 2026-09-03.
Artificial Analysis (2026). Gemini 3.8 Flash release. Artificial Analysis, independent benchmarking service. https://artificialanalysis.ai/models/releases/gemini-3-8-flash Verified 2026-09-03.
Cost-per-task figures are modeled by Capital & Compute from the published rates above, using the task profiles on the cost-per-task calculator: cost equals cached tokens times the cache rate, plus fresh input tokens times the input rate, plus output tokens times the output rate. The monthly and cumulative figures assume five multi-step agentic tasks per working day across 22 working days per month.