On October 19 Inception released Mercury 3, which it calls the most capable diffusion LLM (dLLM) to date. Per the vendor, it is more than 40x the size of Mercury 1, outputs 1,068 tokens/s, expands context from 260K (Mercury 2.5) to 1M tokens, and beats GPT-6 Luna and Claude Haiku 5.5 (both medium) on GPQA Diamond, IFBench, SWE-Bench Atlas and more. List price is $0.20/M input and $0.70/M output, 50% off at launch. It is live on the Inception API, OpenRouter, Baseten and Vercel AI Gateway with an OpenAI-compatible API. All scores are vendor-reported and not yet independently reproduced.
Key Takeaways
- ✓Speed: 1,068 tokens/s output; the vendor table lists GPT-6 Luna at ~130 and Haiku 5.5 at ~240 (vendor figures).
- ✓Scores: GPQA Diamond 89.23, IFBench 76.84, SWE-Bench Atlas 51.08, all above Luna and Haiku 5.5; but DeepSWE v1 at 46.31 trails Haiku 5.5 (51.62), and MRCR v2 also trails Haiku.
- ✓Scale and context: more than 40x the parameters of Mercury 1, with a 1M-token window (260K on Mercury 2.5).
- ✓Price: $0.20/M in and $0.70/M out list, 50% off at launch ($0.10/$0.35), cached input 90% off; new API keys get 100M free tokens.
- ✓Access: Inception API, OpenRouter, Baseten and Vercel AI Gateway, OpenAI-compatible as a drop-in replacement.
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and Pain Point
Mainstream LLMs are autoregressive: they emit one token at a time, and bigger means slower. For browser agents, real-time search and voice support, where every step waits on the model, latency compounds, pushing teams toward smaller, weaker models. Inception has bet on diffusion language models since Mercury 1 in June 2025; Mercury 3, its fourth release, aims to show diffusion models can scale without slowing down.
Architecture Highlights
A diffusion LLM drafts a whole answer and then refines many tokens in parallel over several passes with a Transformer, instead of writing left to right. Inception says Mercury 3 is more than 40x the size of Mercury 1 yet still runs at 1,068 tokens/s on NVIDIA GPUs, is 3x more intelligent on standard benchmarks than Mercury 1 mini, and has a 1M-token context. The exact parameter count was not disclosed.
Benchmarks
All vendor figures, against medium tiers: GPQA Diamond 89.23 vs GPT-6 Luna 86.36 and Haiku 5.5 84.51; IFBench 76.84; SWE-Bench Atlas 51.08 vs 33.33 and 32.26; BrowseComp+ 55.18. It trails on some: DeepSWE v1 46.31 vs Haiku 5.5 at 51.62, and slightly below Luna on AA-LCR and AA-Omniscience. Partner TinyFish reports 50% completion on 50 live Mind2Web tasks. No independent reproduction yet.
Developer Hands-On Guide
Mercury 3 is OpenAI API compatible, so switching means changing the base URL and model name; it is also on OpenRouter as inception/mercury-3 and on Vercel AI Gateway. Launch pricing is $0.10/M in and $0.35/M out, with 100M free tokens for new keys. It fits high-frequency, low-latency subtasks inside agent loops; for complex long-horizon coding, compare on your own evals before switching.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.