MindPattern
← The Wire

Public story · 2026-07-09 · high

Cognition Ships SWE-1.7: Near-Frontier Coding at ~$2/Task, But Devin-Only

Cognition launched SWE-1.7 on July 8, its most capable coding model, RL post-trained on Kimi K2.7 Code inside the real Devin harness for a 12.2-point FrontierCode 1.1 jump over the base.

Why now: This entered the 2026-07-09 research corpus from Cognition and has enough source evidence to become a public Rabbit Hole story.

Confidence
high
Sources
1
Redaction
passed

Story

Cognition launched SWE-1.7 on July 8, its most capable coding model, RL post-trained on Kimi K2.7 Code inside the real Devin harness for a 12.2-point FrontierCode 1.1 jump over the base. It scores 42.3% on FrontierCode 1.1 (vs GPT-5.5's 43.0% and Opus 4.8's 46.5%) and 81.5% on Terminal-Bench 2.1, running at ~1000 tok/s via Cerebras for roughly $1.97/task. The catch for builders: it's exclusive to Devin with no direct API.

Related stories

Each link below shares sources, entities, or timing with this story.

  1. Same source

    Grok 4.5 is Opus-class at 40% of the price, and the independent evals mostly agree

    Cite the same source (Cognition).

  2. Semantically similar

    Meituan open-sources 1.6T coding model built on Chinese chips

    Covers closely related ground (similarity 0.69).

  3. One transformer layer can match full RL fine-tuning

    Covers closely related ground (similarity 0.64).

  4. Anthropic Makes Sonnet 5 the Default in Claude Code

    Covers closely related ground (similarity 0.63).

  5. Claude Code was quietly tagging requests from Chinese AI gateways before Anthropic pulled the code

    Covers closely related ground (similarity 0.62).

  6. Anthropic Brings Claude Fable 5 Back Online July 1

    Covers closely related ground (similarity 0.61).

  7. Anthropic launches Claude Science, reportedly lands Jumper

    Covers closely related ground (similarity 0.61).

  8. Cloudflare will let you charge AI agents per request for anything you host

    Covers closely related ground (similarity 0.60).

Source trail

Cognition

Claim evidence

  1. Cognition launched SWE-1.7 on July 8, its most capable coding model, RL post-trained on Kimi K2.7 Code inside the real Devin harness for a 12.2-point FrontierCode 1.1 jump over the base. It scores 42.3% on FrontierCode 1.1 (vs GPT-5.5's 43.0% and Opus 4.8's 46.5%) and 81.5% on Terminal-Bench 2.1, running at ~1000 tok/s via Cerebras for roughly $1.97/task. The catch for builders: it's exclusive to Devin with no direct API.
    Finding 15754Source

Provenance

Canonical issue
2026-07-09
AI generated
yes
Story unit
15754
Labels
source-backed