PRODUCTS

Mercury 3 makes the case for a new kind of AI

While the AI industry has bet heavily on LLMs and the massive data centers needed to scale them, one of the neolabs betting on a contrarian AI just released another new model that challenges the incumbent thesis. 

On Monday, Inception announced Mercury 3, its latest flagship model that's a diffusion LLM that uses different technology than traditional LLMs, generates text much faster at a lower cost, and is now matching low- and mid-tier models from Anthropic and OpenAI on performance and quality. Inception also reported that it implemented swarms of Mercury 3 agents for recursive self improvement where they increased the overall speed of the model by 16% on top of the upgrades made by the Inception team itself.

"Mercury 3 is the most capable diffusion-based language model on the market," Inception CEO Stefano Ermon told The Deep View. "It's very intelligent. It's comparable to the Luna and Haiku models, but it's way faster. It's getting 1,000 tokens per second. It's almost instant in the way you get answers from these models."

Mercury 3 costs $0.10 per million input tokens and $0.35 per million output tokens. Compared to the leading models from Anthropic's Haiku 5.5 and OpenAI's GPT-6 Luna, that's the same price for input tokens and a 35% discount on output tokens. However, that's based on the launch price for Mercury 3, which is 50% off the eventual standard price. Inception hasn't said how long the launch promotion period will last. But Ermon told The Deep View that the company is working on developments that could allow it to make the launch price permanent in the weeks ahead. 

However, thinking of Mercury 3 as a low-cost model isn't the right framing. Anthropic and OpenAI offer cut-rate prices on their budget models so that they can offer a low-cost option while charging steep premiums for their best models. Mercury 3 is Inception's flagship model and its number one benefit is speed more than cost. Inception claims that Mercury 3 performs 4x to 8x faster than Haiku 5.5, GPT-6 Luna, and GLM-5.3 Flash while performing better than all of them on agentic benchmarks. 

The way Mercury 3 pulls this off is by parallelizing inference. Conventional LLMs generate text one token at a time in sequence. That's an autoregressive process, meaning that each token is based on the ones before it. Meanwhile, Mercury 3 can generate and refine multiple tokens at the same time. This takes advantage of the enormous parallel processing capabilities of GPUs in a much faster and more efficient way. That's what allows Inception's models to produce responses faster and at a lower cost. Diffusion models are used by many of the leading labs for image and video generation. It's the application of the technology to text that's the breakthrough here.

Mercury 3 is now using 40x as many parameters as Mercury 1 (released in February 2025), but the Inception team isn't saying how many total parameters are in the model. That likely means it's a smaller number and Inception doesn't want people to compare apples to oranges and underestimate the model. Still, Mercury 3 shows that diffusion LLMs are advancing rapidly and could match traditional frontier LLMs in the next year. That's likely why Google DeepMind is quietly working on the same technology as well, releasing DiffusionGemma in June. By the way, it uses 26 billion parameters (miniscule compared to the 1 trillion parameter frontier models) but also delivers similar breakthroughs in speed to Inception's Mercury models.

Mercury 3 calls into question the current approach of most of the leading AI labs, which have bet their architecture on building bigger LLMs and then count on faster hardware (and more of it) from Nvidia and other chipmakers to deliver improvements over time. If Inception is right about diffusion LLMs, then it would have big implications. The dilemma facing the AI industry is that improving model intelligence requires additional compute, and lots of it. Upgrading reasoning and increasing the number of agentic steps can typically increase costs and response time. So Inception's breakthrough on diffusion LLMs is all about faster AI response times and more efficient use of expensive GPU hardware, for now. The question remains whether it will be a niche use-case for more real-time AI applications or a bigger long-term challenger to the cutting edge frontier models.

TOGETHER WITH HYPHENATE

Your pricing evolved. Your revenue recognition needs to keep up.

Per-API charges. Prepaid credits. Fees tied to successful outcomes.

Usage-based pricing creates revenue recognition questions that can slow your close, and leave your team searching for answers.

Hyphenate’s Usage-Based Revenue Recognition Playbook tackles six judgment calls, from gross-versus-net reporting to unbilled usage and credit breakage. Each comes with a realistic scenario, a decision tree, journal entries, and an operational fix. Written by a CPA and reviewed by Big Four partners, the guide cites specific codification subparagraphs and cross-checks positions against two Big Four handbooks.

Get the playbook.

AI TRAINING

Delegate a task to an agent, get back a pull request

Most AI coding agents run on a laptop with the developer’s credentials and no record of what they touched. Coder Agents moves the loop into the Coder control plane. You describe a task, the agent provisions a workspace from a template you already use, does the work, and opens a pull request. No keys or agent software in the workspace, and every action tied to your identity.

How to do it:

1. Open a chat. Sign in to your Coder deployment and open Agents. No deployment? Try this on a live environment at coder.com/learn, or self-host with the install docs.

2. Describe the task and choose Plan first. Write it like a handoff to a teammate, name the repo, and pick Plan first in the chat input menu. The agent chooses a template, provisions a workspace under your identity, and proposes a plan before editing anything.

3. Approve the plan, then keep talking. Reply to adjust the plan, then click Implement when it looks right. The agent starts editing, testing, and committing. Follow-ups queue while it works, so add context any time.

4. Review the pull request. Open it in VS Code, JetBrains, or Cursor, or connect your IDE to the workspace the agent used. Every action is logged against your identity.

Nothing about the task changed, only where it ran. The loop stayed in your Coder deployment, the workspace came from a template your platform team already governs, and the audit trail starts at the prompt instead of ending on a laptop.

CULTURE

AI's philanthropy push is also about power

Public sentiment toward the AI industry is broadly negative. Now, the companies are leaning into public-good initiatives to try and change the narrative.

On Sunday, Anthropic announced the Anthropic Public Benefit Fund, a new nonprofit arm that, subject to stockholder approval, will receive 5% of Anthropic's shares, with an additional 2% added at the board's discretion for every 50% increase in the company's value above the a set baseline, according to the blog post. Anthropic says the Fund is intentionally structured in a way that, the larger it gets, the more it can contribute to society.

While the company already operates as a public benefit corporation, or a company built to balance profit with public good, Anthropic says this new arm will have increased flexibility and scale to take actions such as making large donations to traditional nonprofits. 

The fund will also work in tandem with the public good initiatives it already has in place, including:

• Committing  $200 million to the Anthropic Economic Futures Research Fund that focuses on research of the economic impacts of AI;

• The companies' co-founders' pledge to donate 80% of their equity holdings; 

• And the equity donation match program that allows employees to pledge equity which Anthropic matches. Half of its employees participate in it currently.  

Meanwhile, Elon Musk, CEO and founder of SpaceX (and its AI arm known as SpaceXSI)  announced on X that next year, he plans to give one of his SpaceX shares to "every kid in America." He also claimed that it would be the first of many, offering no additional details. 

As CNBC reported, Musk currently owns about 4.8 billion SpaceX shares, and there were 72 million children under 18 in the US as of July 1, 2025, according to census data. Musk would be, therefore, giving away about $10.5 billion worth of shares, using SpaceX's closing stock price on Friday. As this announcement is limited to an X post, and Musk has a track-record for not following through on philanthropic promises, there is no guarantee that it will actually come to fruition. Yet, even if it does, it would still be lapped by Anthropic's contributions, which will likely come out to the $75 billion to $100 billion range, according to Reuters.

There is no denying that, at face value, these are generous donations whose outcomes would be beneficial, and certainly better than no contribution at all. It is also worth noting, however, what these companies stand to gain from taking actions of this sort. Prioritizing public perception is becoming increasingly necessary for them to carry out the plans they need to maximize profits. The data center buildout, for instance, depends in large part on local and federal policymakers, whose decisions are influenced by how they are perceived by the public. Without sustained goodwill, that support may not materialize. Securing favor with those in a position to influence those decisions matters just as much. Anthropic is in an especially vulnerable position: the company could go public as early as November, in what would potentially be the largest IPO of a public benefit corporation ever, and markets are moved by investor perception.

TOGETHER WITH BOX

Reimagine enterprise AI at BoxWorks — Nov. 5-6

Unlocking enterprise AI begins by connecting agents to where your business knowledge lives: your content. Box securely connects unstructured data to the tools your teams use daily — automating workflows, removing bottlenecks, and turning content into repeatable business outcomes.

Join BoxWorks Nov. 5-6 in San Francisco to:

• Hear from AI visionaries like Box's Aaron Levie, Nvidia's Jensen Huang, and Intel's Lip-Bu Tan

• Hone your skills around AI innovation and agents

• Get a first look at the latest in Content + AI

Save your spot with code DeepView50 for 50% off your registration

PRODUCTS

China's open models threaten US AI's pricing power

As China continues to dominate the open AI ecosystem, competition between the country's labs is growing more heated. 

On Monday, Chinese AI lab MiniMax launched an updated version of M3.1 Flash, its latest model which it claims delivers a "generational leap" in agentic tasks, coding, computer use and cybersecurity tasks. The multimodal model, codenamed "space bunny," also features a one million token context window, speeds up online inference threefold and is 70% cheaper than its previous iteration. 

In its announcement, the company said that M3.1 Flash "represents our ongoing exploration of the balance between efficiency and capability." 

The company claims that M3.1 Flash sweeps the board in a number of benchmarks, including for software engineering, reasoning, long-running tasks, coding and more, outperforming both Chinese competitors like DeepSeek, Moonshot's Kimi, and Z.ai's GLM. Minimax also claims its models outperform OpenAI's GPT-5.6 Sol and Anthropic's Opus 5 in the same metrics. 

Additionally, the company claims its model is one of the most cost-effective on the market, running at 15 cents per million input tokens, 60 cents per million output tokens and less than 2 cents per million cached input tokens. For reference, that's a fraction of the price that model providers like OpenAI, Anthropic and Google are selling even their most affordable models for. 

However, Minimax isn't the only Chinese lab gaining recognition in recent days. Last week, DeepSeek released V4.1-Flash-1017, its latest model that similarly sports faster task completion, strong performance across benchmarks and improved costs. The model is a refresh of a DeepSeek model that was initially launched in September. Over the weekend, DeepSeek's model snagged the top score on Code Arena's WebDev leaderboard for open models, and 12th place for overall models. Additionally, leaderboards from Artificial Analysis show DeepSeek's model outperforming MiniMax M3.1-Flash in some benchmarks.

Open models could completely shift the economics of AI, offering an attractive price point to developers and enterprises that are quickly growing fatigued of the price tags that frontier labs are charging for their proprietary APIs. Updates like these prove that open models are quickly closing the performance gap with their proprietary, closed-source counterparts. While some of this development is happening in the US, such as Nvidia's Nemotron models, Reflection's Beam, and Meta's recent release of Muse Spark Open, the vast majority of open source AI innovation is concentrated in Chinese firms. And the models that these labs put out come with a few caveats, including thinner safety and security measures and accusations of model distillation.  

LINKS

• AI agents attempted 1,200 break-ins, compromised over 100 sites

• Google Deepmind agents beat Minecraft autonomously on Gemini 4 Argon

• Tenstorrent, an AI chip designer, raises $1.5 billion, $6.2 billion valuation

• Compute startup DoubleAI launches AI category to cut computing costs

• Disney seeks Character.AI assets via bankruptcy amid child safety lawsuits

• Microsoft treats AI agents as insider threats with new tools

• Reason: A simple, state-of-the-art coding agent.

• Exa 3.0: The company's latest model with its "largest ever improvement to search quality."

• ChatGPT Slides: Users can now work together in real time and make edits right on slides. 

• Ming-3.0: Ant Group open sourced five models across three speech families.

• Fonzi AI: ML Engineer

• Kaizen Equity Partners: AI Operations Associate

• Moring AI: Solutions Architect

• SB Telecom America Corp: AI/ML Engineer, AI Infrastructure

GAMES

A QUICK POLL BEFORE YOU GO