Accelerate inference, model shaping, and pre-training on a research-optimized platform.
San Francisco, CA
Joined November 2022
- Tweets3.2K
- Following279
- Followers64.9K
- Likes2.5K
Kimi K3.1 Pro has arrived, and the weights are dropping tomorrow.
Tomorrow, hear @Kimi_Moonshot 's Feihu Tang explain the architecture and decisions behind it.
He joins Austin Silveria, Kevin Cui and Zain Hasan from Together AI for a deep dive into long-running agents and production inference.
🗓 Wed, Nov 4 · 9 AM PT / 12 PM ET Register → luma.com/together-1w4x
Europe’s largest private company just made open models its default for coding.
@Revolut: ~2,000 engineers, 1/10th the cost, 70% faster lead time.
Proud to be part of the infrastructure powering open models in production 🚀.
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
“Token maxing without cost maxing.” That’s the real goal.
@vipulved at #TCDisrupt on why the smartest teams are swapping closed models for smaller, faster ones.
youtube.com/watch?v=niojbz9f…
training and generation stay aligned on together post-training
that means the reward signal you designed is the one the model actually learns from, and the behavior carries through to serving
together.ai/post-training
one benefit of an open model stack: every task doesn’t have to go through the same model
well-specified work can run on smaller, faster models, while harder or more ambiguous tasks benefit from larger ones
match the model to the work instead of defaulting to the biggest one
Cool use case for open models!
If you want your apps to be more unique, generate your own fonts!
These only cost $0.02 each with DeepSeek V4.1 Flash 1017 on @togethercompute.
Open sourcing all 36 fonts on Monday!
agentic rl means training across long, multi-turn interactions with tools
together post-training supports tool calls, async generation + the environment you already use
you define the environment and rewards. we run the training + generation
together.ai/post-training
🇺🇸 Introducing geo-specific serverless inference endpoints, starting with seven leading open models processed 100% in the U.S.
Switch your base URL to us-api.together.ai/v1
Same API key, same model IDs. 1.25x standard rates.
Learn more: togetherai.link/ZA0OzLW
jack & jill used together post-training to train a model that predicts which candidates a hiring manager will want to meet
in offline evals, they matched a closed-source model’s quality on an open-source model that’s significantly cheaper to serve
together.ai/post-training
Speech-to-speech models have traditionally sacrificed intelligence. Phonic-1 from @phonic_co is the most advanced speech-to-speech model for agentic tasks, running highly challenging calls across six continents.
Trained and served on Together AI. Congrats to the Phonic team!
Introducing Phonic-1, the most advanced speech-to-speech model for agentic tasks — from @phonic_co, the speech-to-speech model company.
We’ve had a breakout year and raised a $27.5M Series A led by @firstround with participation from @Lux_Capital, @felicis, and Definition.
Together AI retweeted
Rain couldn't stop the builders.
Great conversations at our Step 5 Preview Launch Night with @togethercompute!
From long-horizon agents and tool-calling reliability to inference costs and what AI breakthroughs mean for the next wave of startups.
Huge thanks to our panelist Dan Fu (@realDanFu · @togethercompute), Mar Hershenson (@MarHershenson · @pearvc), Saoud Rizwan (@sdrzn · @cline), and Joey Wang (@harvey), and everyone who joined us!
More to come.
long agent rollouts can get expensive fast
together post-training uses per-gpu pricing, so running more rollouts in parallel brings the cost of each one down
together.ai/post-training
together gpu clusters can now take a node from gpu fault to healthy again automatically
health checks identify the failure, the platform chooses the lightest repair, and you decide what can run unattended, down to individual xid codes
here’s how it works
together.ai/blog/auto-repair…
Two companies, one goal: open models that work for builders at any scale.
@jmorgan of @ollama and @vipulved of Together AI on where open models win, how to spend tokens wisely, and what developers want next.
📽️: youtube.com/watch?v=ZAGVED3p…
the open model stack is becoming composable
model → inference → gateway/router → harness → tools
each layer can evolve independently, so you can swap in better models, providers + tools as the ecosystem moves
we broke down the full stack ↓
together.ai/blog/the-open-so…
minimax m3.1 flash beats opus 5.5 on agentic automation at ~27x lower cost
the model has a 1m context window + vision support, and is built for coding, computer use + cybersecurity
We are LIVE with @AIatMeta to dive into Muse Spark Open!
Everything you need to know about Muse Spark Open: A conversation with Meta and Together AI teapawt.org/i/broadcasts/1MJgNbkXm…
Everything you need to know about Muse Spark Open: A conversation with Meta and Together AI teapawt.org/i/broadcasts/1MJgNbkXm…
post-training gets better the faster you can iterate
multiple lora experiments can share the same gpu capacity on together post-training, so you can test more ideas without spinning up new infrastructure for every run
together.ai/post-training
