Mercury 3 is live on OpenRouter!
@_inception_ai's new diffusion model generates 1,000+ tok/s with 1M context, 4x more than Mercury 2.5. Inception reports it beats GPT-6 Luna and Haiku 5.5 on IFBench, GPQA Diamond, and SWE-Bench Atlas.
50% off at launch: $0.10/M in, $0.35/M out
@_inception_ai's new diffusion model generates 1,000+ tok/s with 1M context, 4x more than Mercury 2.5. Inception reports it beats GPT-6 Luna and Haiku 5.5 on IFBench, GPQA Diamond, and SWE-Bench Atlas.
50% off at launch: $0.10/M in, $0.35/M out
Oct 19, 2026 · 4:54 PM UTC
48
33
743
46,048
@OpenRouter @_inception_ai That pricing for 1M context and speed is wild
0
0
1
34
@OpenRouter @_inception_ai Big fan of this.
0
0
1
28
@OpenRouter @_inception_ai Insane throughput with massive memory window at killer debut pricing love this
0
0
0
60
@OpenRouter @_inception_ai 1,000 tok/s with 1M context is insane throughput great to see Mercury 3 live and making diffusion models so accessible
0
0
0
74
@OpenRouter @_inception_ai I'm on Haiku 5.5 right now. Guess I know what I'm benchmarking this week.
0
0
5
994
@OpenRouter @_inception_ai I was waiting to see it pop up. It shows up as released 3d ago. That's why I missed it initially.
0
0
3
435
@OpenRouter @_inception_ai What can it be used for?
1
0
1
673
@OpenRouter @_inception_ai @grok leur api comment on recoit les tokens dans le harness ? genre on stream le thinking, reponse etc mais faut adapter pour les modèles diffusion? ou bien eux meme font en sorte de rendre le truc compatible ? et du coup ya pas de stream ?
2
0
1
807
@OpenRouter @_inception_ai Impressive speed and strong benchmark performance
0
0
0
79
@OpenRouter @_inception_ai 1000 tok/s with 1M context at that price is wild beating Luna and Haiku on IFBench/GPQA while launching at 50% off
0
0
0
28
@OpenRouter @_inception_ai 1M context at 4x speed is a BANGER for SI workflows.
0
0
3
293
@OpenRouter @_inception_ai tonight. with reasoning off it's really quick, 0.4s a move in bullet chess and never lost on time. with reasoning on it'll sometimes think for over a minute on one answer. business tasks (survey analysis, decks, ad budgets) came out near the bottom of our tables
0
0
1
200
@OpenRouter @_inception_ai Love seeing Mercury 3 on OpenRouter with that launch pricing 1,000+ tok/s and 1M context for $0.10/M is an amazing deal!
0
0
1
63
@OpenRouter @_inception_ai Does the speed hold near 1M context, or fall off deep in?
0
0
3
282
@OpenRouter @_inception_ai at 1000 tok/s my test suite becomes the slow step in every loop, the model would just sit there waiting for maven ☕
0
0
3
478
@OpenRouter @_inception_ai When I want a fast, really fast, work horse model I turn to Mercury 2.5. Super excited to see Mercury 3 drop!
0
0
1
5
@OpenRouter @_inception_ai Maybe it is, but not on my end :) lets give it an hour and try again
0
0
1
316
@OpenRouter @_inception_ai Insanely fast massive context and half price at launch is a steal
0
0
1
26
@OpenRouter @_inception_ai 1M context on a diffusion lm surprised me, every denoising step re-attends the whole prompt. guessing @_inception_ai caches it as a frozen prefix block
0
0
1
284
@OpenRouter @_inception_ai @_inception_ai was and always will be a fan of mercury models , switching to it today! yall do an amazing job
0
0
0
16
@OpenRouter @_inception_ai 40x+ more params than mercury 1 and still ~1,000 tok/s is the real headline. an autoregressive model that much bigger would crawl at batch 1, diffusion gets away with it by refining many tokens per forward pass. what does TTFT look like on long prompts?
0
0
0
@OpenRouter @_inception_ai $0.10/M in and $0.35/M out at 1,000+ tok/s makes Mercury 3 an easy pick for the fast inner steps of a coding agent. Are people routing to it as the primary model or mostly as a cheap fallback?
0
0
0
1
@OpenRouter @_inception_ai Great launch Mercury 3 looks incredibly fast and powerful amazing performance and pricing
0
0
0
28
@OpenRouter @_inception_ai 1000 tok/s at that price is crazy cheap. definitely gonna swap the model id in my config and run some tests after work
0
0
0
33
@OpenRouter @_inception_ai Mercury 3 now live on OpenRouter Inception's diffusion model does 1,000+ tok/s, 1M context (4x Mercury 2.5), beats GPT-6 Luna & Haiku 5.5 on IFBench, GPQA Diamond, SWE-Bench Atlas. Launch pricing 50% off: $0.10/M in, $0.35/M out
0
0
0
26
@OpenRouter @_inception_ai Day-one OpenRouter support is exactly why I run half my evals through you folks. Swapping Mercury 3 into my speed benchmark tonight, curious whether that 1M context holds up under real load.
0
0
0
123
@OpenRouter @_inception_ai Mercury 3 is on OpenRouter: 1,000+ tok/s and a 1M context, 4x more than Mercury 2.5, at $0.10/M in and $0.35/M out with 50% off at launch. picture the whole answer developing at once like a polaroid while the model next to it is still typing "great question."
0
0
1
200