Meta and Google have landed fresh models in the middle of an already crowded AI release cycle. Meta introduced Muse Spark 1.3, while Google pushed out Gemini 3.8 Flash, giving developers two new options built around stronger reasoning, coding and agentic capabilities without pushing pricing sharply higher.
The launches also show where the AI race is moving. Raw intelligence still matters, but cost, speed and the ability to complete real tasks are becoming just as important. Meta is leaning heavily into that combination with Muse Spark, while Google is trying to regain momentum by iterating on Gemini at a much faster pace.
Meta Pushes Muse Spark 1.3 Further Into Frontier Territory
The model has been designed to maintain multiple tasks inside the same conversation, use external tools, gather missing information and revise its approach when something goes wrong. That matters because AI models are increasingly expected to do more than answer a single question. Developers now want systems that can work through complicated tasks over several stages without losing track of the original objective.
Meta also claims the model is more efficient than Muse Spark 1.2. According to the company, Spark 1.3 can complete certain coding workflows while using fewer tool calls and fewer tokens. That kind of improvement may not attract the same attention as a benchmark headline, but it could have a direct effect on how much businesses pay when running AI agents at scale.
Muse Spark 1.3 Makes Cost Part of the Competition
One of the strongest parts of Meta’s pitch is cost. Muse Spark 1.3 keeps its standard API pricing at $1.25 per million input tokens and $4.25 per million output tokens, while cached input costs considerably less. Meta is not simply trying to compete by cutting the headline token price. Instead, the company is also trying to reduce the amount of work the model needs to perform before completing a task.
That distinction could become important as agentic AI expands. A traditional chatbot response may only require one prompt and one answer, but an AI agent can make repeated tool calls, inspect information, revise plans and generate much larger amounts of output. A model that finishes the same job with fewer steps can end up being cheaper even when its listed API price remains unchanged.
Mark Zuckerberg described Muse Spark 1.3 as delivering frontier-level performance at a price that is almost too cheap to measure. The language is deliberately aggressive, but it reflects Meta’s broader strategy of trying to compete not only on intelligence, but on how much usable intelligence developers can get for each dollar.
Muse Spark 1.3 Posts Strong Benchmark Results
Muse Spark 1.3 also drew attention because of its performance on independent AI benchmarks. At launch, the Max version reportedly reached a score of 62 on Artificial Analysis’ Intelligence Index, placing it close to the top of the leaderboard under the methodology being used at the time. That result helped reinforce Meta’s claim that smaller and more affordable models are moving much closer to the industry’s most capable systems.
Benchmark rankings can change quickly, however. Artificial Analysis later updated its Intelligence Index methodology, and Muse Spark 1.3 received a different score under the newer version. That makes the original 62 figure useful as a snapshot of its launch performance rather than a permanent ranking. The larger point remains that Meta is closing the performance gap while keeping the model relatively inexpensive to run.
Watermelon Could Be Meta’s Bigger AI Move
Muse Spark 1.3 may only be the beginning of Meta’s next wave of AI releases. Zuckerberg has already teased a larger model known internally as Watermelon, while Meta has also indicated that an open-weight version of Muse Spark is coming. The company has not provided a firm public release date for the larger model, but the teaser has already raised expectations around what Meta plans to release next.
Watermelon could become a much bigger test of Meta’s AI strategy. Muse Spark shows that the company can compete strongly in the smaller, cost-efficient model category. A larger model performing near the absolute frontier would put Meta in more direct competition with OpenAI, Anthropic and Google at the top end of the market.
An open-weight Muse Spark release could also widen Meta’s appeal among developers. Open weights would give companies more control over deployment, fine-tuning and infrastructure, particularly for organisations that do not want to rely entirely on hosted APIs.
Google Responds With Gemini 3.8 Flash
Google introduced Gemini 3.8 Flash on the same day, continuing what has become a rapid sequence of Gemini updates. The company describes the model as its strongest Flash release so far for reasoning, coding and agentic tasks. It arrives only weeks after Gemini 3.7 Flash, showing how quickly Google is now cycling through new versions of its models.
Gemini 3.8 Flash is designed for developers who want stronger performance without moving to a slower or significantly more expensive flagship model. That gives Google a similar objective to Meta: deliver enough intelligence for demanding workloads while keeping latency and cost low enough for large-scale use.
The model is also positioned heavily around software development and multi-step reasoning. Those are two of the most competitive areas in AI right now, particularly as companies build coding agents that can inspect repositories, write software, test changes and use external tools with less human supervision.
Gemini 3.8 Flash Keeps Pricing Aggressive
Google has kept Gemini 3.8 Flash at the same introductory pricing level as Gemini 3.7 Flash, with input priced at $0.75 per million tokens and output at $3.75 per million tokens. That puts it below Muse Spark 1.3 on the published input and output rates, although real-world costs can still vary depending on how many tokens and tool calls each model needs to complete a task.
The model also supports a large context window and multimodal inputs, making it suitable for workflows involving text, images, audio and video. Google is making it available across its Gemini developer ecosystem, including the Gemini API and Google AI Studio, which gives the company a broad distribution advantage among developers already using Google infrastructure.
Keeping the price steady while improving performance is important for Google because developers increasingly compare models on more than benchmark scores. A technically stronger model can still lose adoption if another option completes the same work faster or more cheaply.
Google Still Wants a Return to the Frontier
Despite the progress with Gemini 3.8 Flash, Google has made it clear that it does not consider the race finished. DeepMind leadership has acknowledged that Gemini currently sits slightly below the absolute frontier in some areas, while also making clear that returning to the top remains a priority.
That puts additional pressure on whatever Google releases next. Gemini 3.8 Flash looks more like a strong step forward than the final answer. If Google’s next major Pro-class model does not arrive until Gemini 4, competitors have more time to establish stronger positions with their own flagship systems.
Google once had a clear reputation for combining capable AI with aggressive pricing. Meta is now challenging that position directly, which makes the next stage of the Gemini roadmap even more important.
Coding and AI Agents Are Becoming the Main Battlefield
The similarities between Muse Spark 1.3 and Gemini 3.8 Flash reveal a broader shift across the AI industry. Both models emphasize coding, multi-step reasoning, tool use and agentic work. Those features are becoming much more important than simple chatbot performance because businesses increasingly want AI systems that can actually perform tasks rather than just provide information.
This changes how developers evaluate models. A slightly higher benchmark score may not matter if the model requires significantly more tokens, produces more errors or takes longer to finish a job. Efficiency, reliability and the ability to recover from mistakes can be just as valuable as raw reasoning power.
That is why Meta’s cost story matters so much. It is also why Google’s decision to keep Gemini 3.8 Flash pricing low is significant. Both companies are trying to make advanced AI practical enough to use repeatedly, not just impressive enough to demonstrate.
Why Meta Muse Spark 1.3 Matters
Meta Muse Spark 1.3 matters because it strengthens the argument that frontier-like performance is becoming available at much lower prices. Meta is no longer competing only through open models or social platforms. It is increasingly positioning itself as a serious model provider for developers building complex AI systems.
The bigger question is what happens next. If Watermelon delivers another large jump in performance, Meta could move much closer to the absolute frontier. If Muse Spark is released with open weights as planned, developers may also gain a compelling alternative to fully closed commercial models.
Google’s response shows that it understands the pressure. Gemini 3.8 Flash improves performance while keeping pricing aggressive, but Google still needs to prove that its next flagship release can reclaim a stronger position at the top of the AI market.
The result is a model race that now moves in weeks rather than years. Meta is getting cheaper and smarter. Google is iterating faster. Developers are getting more capable tools at lower prices, and the next model announcement is unlikely to be far away.
Sources
Meta AI Research — Introducing Muse Spark 1.3
Google — Introducing Gemini 3.8 Flash and Gemini 3.8 Flash Cyber
