# DeepSeek V4.1 Flash, Devin's SWE-2 and a petabyte of DNA predictions

> Week of 7 September 2026: DeepSeek opens V4.1 Flash under MIT, Cognition's SWE-2 nears Fable 5.1 at lower cost, and DeepMind publishes AlphaGenome Atlas.

Published 13 Oct 2026. Coverage: 7 Sept 2026 to 13 Sept 2026. By Ali Asif.

## Items

1. **DeepSeek V4.1 Flash** (DeepSeek, Open weights): A 552B mixture-of-experts model under MIT with native vision, 1M context and a new encoder-decoder design that uses 8B active parameters for input and 16B for output. [Primary source](https://www.deepseek.com/en/news/deepseek-v4-1-flash/)
2. **SWE-2** (Cognition, Agents and coding): Devin's new coding model scores 50.0% on FrontierCode 1.1 Main, within a point of Fable 5.1 and 64% cheaper, according to Cognition. [Primary source](https://cognition.com/blog/swe-2)
3. **GPT Image 2.5 Flare** (OpenAI, Image, video, world): OpenAI's fastest image model in the 2.5 line, available in the Image API and as a tool inside the Responses API. [Primary source](https://developers.openai.com/api/docs/models/gpt-image-2.5-flare)
4. **Nex-N2.5-Max** (Nex AGI, Open weights): A 1.6-trillion-parameter, text-only agentic mixture-of-experts model with 49B active, a 1M window and Apache 2.0 weights. [Primary source](https://huggingface.co/nex-agi/Nex-N2.5-Max)
5. **AlphaGenome Atlas** (Google DeepMind, Science): A 1-petabyte precomputed map of predicted effects for about 9 billion possible single-letter DNA changes, free for academic use. [Primary source](https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome/)
6. **Arrow 2** (QuiverAI, Image, video, world): A vector SVG model tuned for speed and cost per asset, with text, image and SVG input and cleaner geometry. [Primary source](https://docs.quiver.ai/api/models)
7. **NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training** (arXiv, Science): Describes a model family that watches its own capabilities and turns that evidence into the next round of training. [Primary source](https://arxiv.org/abs/2609.08183)

A week where open models and a coding product caught up to the flagships on cost, and where biology got a big data release.

## DeepSeek V4.1 Flash

DeepSeek released V4.1 Flash on 10 September under the MIT license. It is a 552B-parameter mixture-of-experts model and the smallest member of a new architecture family, with native image understanding and a 1M-token context. The unusual part is the design: a Causal Encoder-Decoder that activates only 8B parameters when reading input and 16B when writing output. DeepSeek says a smaller KV cache is the main saving, and claims results ahead of its earlier flagship V4-Pro on its own benchmarks. API output can run to 384K tokens.

## SWE-2 from Cognition

The same day Cognition introduced SWE-2, the coding model inside Devin. By the company's account it reaches 50.0% on FrontierCode 1.1 Main, within one point of Fable 5.1, while costing 64% less. It was post-trained from the Kimi K3 2.8T base, scaled reinforcement learning into the multi-trillion-parameter range, and uses an algorithm that trains every reasoning-effort level in one run. Those are Cognition's numbers on a benchmark Cognition publishes, so keep that in mind.

## A genome atlas

DeepMind released the AlphaGenome Atlas on 8 September. It holds precomputed predictions for about 9 billion possible single-letter variants of the human genome, around 1 petabyte, more than 30 times the size of the AlphaFold Database, and free for academic use. For each variant it gives predicted effects across hundreds of human and mouse cell types, plus a single impact score. DeepMind describes using it to spot a rare-disease variant in the DNM1 gene.

## Images and graphics

OpenAI released GPT Image 2.5 Flare and Sunburst on 8 September, with Flare as the fast tier. QuiverAI released Arrow 2 on 7 September, a vector model that outputs SVG. Nex AGI opened Nex-N2.5-Max, a 1.6T agentic model under Apache 2.0.

One more paper: NeoHorse-1 describes a route to recursive self-improvement through agentic post-training.

## What I'd watch next

Whether DeepSeek's encoder-decoder trick shows up in the next large release, and what else gets post-trained from open bases like Kimi K3.

## My take

SWE-2 is the item I'd read twice. Cognition took the Kimi K3 open base and post-trained it into something that, by its own numbers, sits within a point of a flagship at about a third of the cost. That is the open-weights point I make to Saudi buyers: the base model is a commodity input, and the value is in the training on your own workflow.

## Sources

- https://www.deepseek.com/en/news/deepseek-v4-1-flash/
- https://cognition.com/blog/swe-2
- https://developers.openai.com/api/docs/models/gpt-image-2.5-flare
- https://huggingface.co/nex-agi/Nex-N2.5-Max
- https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome/
- https://docs.quiver.ai/api/models
- https://arxiv.org/abs/2609.08183

Canonical: https://heyaliasif.com/ai/2026-09-07-week-in-ai