← The Wire

Models2026-06-29 · source-backed

OpenAI published the GPT-5.6 Preview System Card with the eval data.

Confidence
source-backed
Sources
1
Redaction
redacted

Story

The card (June 26) details OpenAI's strongest safety stack yet, with hardened handling for high-risk cyber and bio requests and repeated-misuse cases. The reason to care isn't the marketing summary. It's that the card includes the actual eval data behind the launch claims, Terminal-Bench 2.1, GeneBench, ExploitBench, which makes it the primary artifact for independently assessing capability instead of taking the press release's word.

Related stories

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

  1. Shared entities / Shared topic / Earlier coverage / Tension

    GPT-5.5 Is Here. It's Fast, It's Expensive, and It Changes How You Prompt.

    Both cover Bench, GPT, OpenAI, Terminal; overlapping topics (capability, gpt-5, openai); earlier Bench coverage from 2026-04-25.

  2. Shared entities / Shared topic / Earlier coverage

    OpenAI shipped GPT-5.6 'Sol' and then handed the keys to Washington

    Both cover Bench, GPT, OpenAI, Terminal; overlapping topics (gpt-5, openai); earlier Bench coverage from 2026-06-26.

  3. Shared entities / Shared topic / Earlier coverage / Tension

    Kimi K2.6: A 1-Trillion-Parameter Open-Weights Model Just Beat GPT-5.4 on SWE-Bench Pro

    Both cover Bench, GPT, Terminal; overlapping topics (capability, gpt-5); earlier Bench coverage from 2026-04-21.

  4. Shared entities / Earlier coverage

    Codex CLI on GPT-5.5 tops Terminal-Bench 2.1 at 83.4%.

    Both cover Bench, GPT, OpenAI, Terminal; earlier Bench coverage from 2026-06-12.

  5. GPT-5.5 powers Codex: 82.7% Terminal-Bench 2.0, native Windows, 2x Pro usage through May.

    Both cover Bench, GPT, OpenAI, Terminal; earlier Bench coverage from 2026-05-11.

  6. Shared entities / Shared topic / Earlier coverage

    GLM-5.2 is the hedge, and the benchmarks say it's a real one

    Both cover Bench, GPT, Terminal; overlapping topics (behind, gpt-5); earlier Bench coverage from 2026-06-21.

  7. Qwen3.5-9B Hits 93.8% on HomeSec-Bench — 4.1 Points Behind GPT-5.4

    Both cover Bench, GPT, OpenAI; overlapping topics (behind, gpt-5); earlier Bench coverage from 2026-03-21.

  8. Shared entities / Shared topic / Earlier coverage / Tension

    OpenAI retired GPT-5.2 and migrated everyone to GPT-5.5.

    Both cover GPT, OpenAI; overlapping topics (eval, gpt-5, openai); earlier GPT coverage from 2026-06-14.

Source trail

Entities

Provenance

AI generated
no
Story unit
2026-06-29-openai-published-the-gpt-5-6-preview-system-card-with-the-eval-data
Labels
source-backed, canonical briefing excerpt