AnswersPublished23 sources
What Inception’s Mercury 3 Launch Claims—and What the Evidence Shows
Inception says Mercury 3 can generate 1,068 tokens per second and outperform GPT 6 Luna and Claude Haiku 5.5 on benchmarks.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Inception says Mercury 3 can generate 1,068 tokens per second and outperform GPT 6 Luna and Claude Haiku 5.5 on benchmarks.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Inception announce with Mercury 3’s October 19 launch, including how its parallel token generation and refinement differ from stand. Article summary: Inception launched Mercury 3 on October 19, 2026, describing it as its most capable diffusion language model and claiming competitive quality at substantially higher speed and lower cost than GPT-6 Luna and Claude Haiku . Topic tags: general, general web, user generated, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks,
Inception launched Mercury 3 on October 19, 2026, positioning it as a faster, more capable diffusion language model. Its headline claims include 1,068 tokens per second on NVIDIA GPUs and benchmark advantages over GPT-6 Luna and Claude Haiku 5.5—but those comparisons should be treated as company claims, not independent validation. 5
20
How Mercury 3 generates text differently
A standard autoregressive language model generates text sequentially, predicting the next token from the tokens already produced. Mercury 3 uses a diffusion approach: it generates multiple token positions in parallel, then refines the draft across successive passes. 6
13
That difference is central to Inception’s speed pitch. Instead of waiting for each token to be produced one after another, Mercury 3 works on a batch of text positions at once. The launch materials report a speed of 1,068 tokens per second on NVIDIA GPUs. 5
20
The launch claims: speed, benchmarks and scale
Inception says Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks while running four to eight times faster. The company’s reported speed comparison puts Mercury 3 at 1,068 tokens per second. 5
20 These figures describe Inception’s claims; they do not establish that the same advantage will hold across workloads or independent testing.
The model is also reported to be more than 40 times the size of Mercury 1, while maintaining speeds above 1,000 tokens per second. Its context window expands from Mercury 2.5’s 260,000 tokens to 1 million. 1
17 Inception’s launch was described as its fourth model release since Mercury 1 launched in June 2025.
29
Early browser-agent result
Inception and coverage of its launch report that Mercury 3 completed 50% of 50 live Mind2Web tasks—25 tasks—and performed better than GPT-6.1-Sol at the same step count. 21
30 This is a specific reported test result, not proof that the model will outperform in every browser-agent setting.
Pricing and availability
The listed prices are $0.20 per million input tokens and $0.70 per million output tokens. A 50% launch discount brings those rates to $0.10 and $0.35, respectively. Cached input tokens are listed at 90% off standard input pricing. 2
5
11
Mercury 3 is available through Inception’s API, Baseten, OpenRouter and Vercel; the API is described as OpenAI-compatible. 24
30
36
What remains to be verified
The main performance figures and comparisons are presented as Inception’s claims. Launch coverage also noted that independent benchmark validation was not available at the time. 11 That makes the announcement useful as a statement of what the company says it has built, but not a settled independent verdict on Mercury 3’s quality or speed relative to competing models.
The clearest takeaway is that Mercury 3 applies parallel generation and iterative refinement to a larger model and a longer context window, with Inception claiming a substantial speed advantage. Independent comparisons will be needed to determine how reliably those headline results translate to real-world use.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "What Inception’s Mercury 3 Launch Claims—and What the Evidence Shows"?
Inception says Mercury 3 can generate 1,068 tokens per second and outperform GPT 6 Luna and Claude Haiku 5.5 on benchmarks.
Sources
siliconangle.comInception says Mercury 3 beats GPT-6 Luna and Claude Haiku 5.5 at up to 8x the speed - SiliconANGLE
inceptionlabs.aiDiffusion at Scale: Introducing Mercury 3 – Inception
gptunnel.ruMercury 3 from Inception: 1,068 Tokens/s and a 1M-Token Context
unite.aiInception Releases Mercury 3, Its Flagship Diffusion LLM