AICoder

AI Coding Live News

First-hand AI coding updates from 𝕏/Twitter frontier labs and researchers, distilled by AI with entity benchmarks.

Explore live on 𝕏 ↗

AI Coding Live News

J
Joël Niklausjoelniklaus·
👀16
🔥90
View on 𝕏 ↗

HF's Joël Niklaus Ships Harness Leaderboard: 28,849 Trials Across 9 Coding Agent Harnesses; Author Reports mini-swe-agent ~51.4% Average

On 2026-11-05, Hugging Face Harness Optimization researcher Joël Niklaus (@joelniklaus) released the Harness Leaderboard: same model, provider, and reasoning effort with only the harness swapped, covering GLM-5.3-Flash, DeepSeek-V4.1-Flash, and Flint on Terminal-Bench 2.1 and DeepSWE—28,849 trials and about $4,746 of inference. Author-reported averages: mini-swe-agent ~51.4% resolve rate; Pi ~49.4% at about $0.29 per solved task. This is an individual research leaderboard, not a Hugging Face company product announcement.

HF's Joël Niklaus Ships Harness Leaderboard: 28,849 Trials Across 9 Coding Agent Harnesses; Author Reports mini-swe-agent ~51.4% Average

Key Takeaways

  • Source: personal post by HF Harness Optimization researcher Joël Niklaus on 2026-11-05T16:04:08Z—not a company announcement
  • Scale (author): 9 harnesses on open models, 28,849 trials, ~$4,746 inference spend
  • Method: pin model/provider/reasoning effort; 3 attempts per task; Terminal-Bench 2.1 and DeepSWE
  • Results (author): averaged mini-swe-agent ~51.4%, openclaw ~27.3%; Pi ~49.4% at ~$0.29 per solved task
  • Use: public HF Space leaderboard + trajectories; Harness Lens documents prompts/tools for each harness
Read full story
Linked Models & Agents:
⚡ GLM-5.3-Flash⚡ DeepSeek-V4.1-Flash🤖 mini-swe-agent🤖 hermes-agent🤖 Claude Code (Anthropic)🤖 codex🤖 opencode🌐 Hugging Face🌐 DeepSeek🌐 Moonshot🌐 Nous Research
#Harness Leaderboard#Joël Niklaus#Hugging Face#Mini swe agent#Terminal-Bench#DeepSWE#Benchmark
A
Aleph Alpha@AlephAlpha·
👀233
🔥99

Aleph Alpha open-sources Kolibri-1: 78B MoE (3.46B active), bilingual DE/EN Apache-2.0; SWE-Verified 66.4

On 3 Oct 2026, Aleph Alpha released open-weight Kolibri-1 (HF, Apache-2.0): 78B MoE / ~3.46B active, bilingual DE/EN, 262k native context (1M extrapolated), reasoning effort + tools. Official card: LiveCodeBench v6 85.9, SWE-Bench Verified 66.4, TerminalBench 2.1 27.7. Serve via aleph-alpha-inference + vLLM.

Aleph Alpha open-sources Kolibri-1: 78B MoE (3.46B active), bilingual DE/EN Apache-2.0; SWE-Verified 66.4

Key Takeaways

  • Shipped 2026-10-03; weights Aleph-Alpha/Kolibri-1 Apache-2.0; blog on aleph-alpha.com
  • 78B total / ~3.46B active; ~78GB FP8; native 262k context (validated to 1M)
  • Official card: LiveCodeBench v6 85.9 · SWE-Verified 66.4 · TerminalBench 2.1 27.7 · HumanEval+ 92.7
  • Serve: aleph-alpha-inference + vllm serve Aleph-Alpha/Kolibri-1 with kolibri1 parsers
  • Sovereign DE/EN on-prem focus — not a Terminal-Bench leader vs denser peers
Read full story
Linked Models & Agents:
⚡ Kolibri-1⚡ Kolibri⚡ Kolibri Origin🌐 Aleph Alpha
#Kolibri#Aleph alpha#Open weight#Moe#Foundation model#German#apache-2.0#Vllm#Coding#Sovereign AI#Long context
C
Cline@cline·
👀66
🔥88

Cline Desktop v0.0.37: About & What's new surfaces; Bedrock OpenAI inference profiles fix reasoningConfig errors

Cline Desktop v0.0.37 (2026-09-26) adds Settings → About (version/channel, check/restart updates, recent release notes, report issue), a one-time What's new dialog after updates, macOS Help→Export Diagnostics, and fixes Bedrock OpenAI inference-profile calls that failed with Unknown parameter reasoningConfig when reasoning effort was set.

Cline Desktop v0.0.37: About & What's new surfaces; Bedrock OpenAI inference profiles fix reasoningConfig errors

Key Takeaways

  • Shipped: desktop-v0.0.37 vs desktop-v0.0.36 on GitHub Releases
  • UX: Settings About page with version/channel, update buttons, release notes links, Report an issue
  • Onboarding: one-time What's new after update; Highlights on About reopens it
  • Ops: macOS Help→Export Diagnostics opens the export directly
  • Bedrock: OpenAI models via us.openai./global.openai. inference profiles no longer fail on reasoningConfig when reasoning effort is set
Read full story
Linked Models & Agents:
⚡ GPT-6🤖 Cline (VS Code Agent)🌐 Cline🌐 Amazon🌐 OpenAI
#Cline Desktop#Bedrock#inference profile#reasoningConfig#What's new#Diagnostics#coding agent
v
vechen@miu21590·
👀88
🔥88
View on 𝕏 ↗

Mid-task Jev in Codex raises/lowers GPT-6 Astra reasoning effort; author reports ~50% Astra cost cut without breaking prompt cache

@miu21590 demoed using Jev inside a running Codex session to raise/lower GPT-6 Astra reasoning effort—more thinking when stuck, less on routine steps—reporting ~50% lower Astra costs and faster runs without breaking prompt caching. Versus the usual pre-task Jev model pick, this wires the decision model mid-run for cost control.

Mid-task Jev in Codex raises/lowers GPT-6 Astra reasoning effort; author reports ~50% Astra cost cut without breaking prompt cache

Key Takeaways

  • Call Jev mid-Codex-run to raise/lower GPT-6 Astra reasoning effort.
  • Author tests: ~50% lower Astra cost, faster, prompt cache intact.
  • Moves Jev from pre-task model pick to in-task cost control.
Read full story
Linked Models & Agents:
🤖 codex🌐 OpenAI
#Codex#Jev#GPT-6#Astra#reasoning effort#prompt cache#Cost
ADSponsored
A
Artificial Analysis@ArtificialAnlys·
👀68
🔥94
View on 𝕏 ↗

Artificial Analysis Intelligence Index v4.3: Terminal-Bench 4.0 + AutomationBench-AA

Artificial Analysis (@ArtificialAnlys) shipped Intelligence Index v4.3: Terminal-Bench upgraded 2.1→4.0 (now mini-SWE-agent harness) and τ³-Banking replaced by AutomationBench-AA (657 private Zapier business workflows). Claude Fable 5.1 and GPT-6 Astra tie at 53; OpenAI holds most of the intelligence-vs-cost frontier across Astra reasoning efforts.

Key Takeaways

  • Benchmarks: Index v4.3 upgrades Terminal-Bench to 4.0 and adds AutomationBench-AA (private set).
  • Weighting: private-task/answer evals rise from 40% to 45%; category weights unchanged.
  • Results: Claude Fable 5.1 and GPT-6 Astra tie at 53; open-weights led by GLM-5.3 / Kimi K3 at 44.
Read full story
Linked Models & Agents:
⚡ GPT-6 Astra⚡ Claude Fable 5.1🌐 OpenAI🌐 Anthropic
#Artificial Analysis#Intelligence Index#Terminal-Bench#AutomationBench-AA#GPT-6 Astra#Claude Fable 5.1#Benchmark
m
melvin@melvindvivas·
👀32
🔥82
View on 𝕏 ↗

JSONL Viewer v1.3.4: nest Codex subagent traces under parent runs with MAIN/SUBAGENT nav

melvin (@melvindvivas) shipped JSONL Viewer v1.3.4: Codex subagent traces now nest under the parent run that spawned them, with MAIN/SUBAGENT labels, parent-child navigation, and per-turn model plus reasoning effort.

Key Takeaways

  • Release: JSONL Viewer v1.3.4 (github.com/donvito/jsonl-viewer).
  • Traces: Codex subagents nest under parent runs with MAIN/SUBAGENT labels.
  • Nav: parent-child jumps; per-turn model and reasoning effort.
Read full story
Linked Models & Agents:
🤖 openai-codex
#JSONL Viewer#Codex#Subagent#Traces#v1.3.4#Debugging
T
Tibo@thsottiaux·
👀56
🔥93
View on 𝕏 ↗

OpenAI Codex lead: GPT-6 Astra on low beats GPT-5.6 Sol on high — dial effort down

OpenAI Codex lead Tibo (@thsottiaux) calibrates Astra reasoning effort: GPT-6 Astra on low already outperforms GPT-5.6 Sol on high. If you were happy with Sol on high, move Astra down to low or medium to avoid over-spending effort.

Key Takeaways

  • Calibration: Astra low already beats Sol high.
  • Guidance: former Sol-high users should try Astra low/medium.
  • Impact: lower default effort can hit quality with less usage burn.
Read full story
Linked Models & Agents:
⚡ GPT-6 Astra⚡ GPT-5.6 Sol🤖 openai-codex🌐 OpenAI
#GPT-6 Astra#GPT-5.6 Sol#Codex#reasoning effort#OpenAI#Tibo
G
GHCopilotLog@GHCopilotLog·
👀44
🔥86
View on 𝕏 ↗

GitHub Copilot App weekly: /goal for local autopilot, MCP deep-link install, Power BI MCP

GitHub Copilot App weekly recap (Aug 29–Sep 5): 1 release and 237 improvements. Highlights include /goal for persistent autopilot objectives in local sessions, MCP install via deep link with a prefilled review form, Power BI MCP in the catalog (hosted remote and local), and configurable reasoning effort for code-review agents. Notes: github.com/github/app/releases.

Key Takeaways

  • /goal sets persistent objectives for local autopilot sessions.
  • Install MCP servers via deep link with a prefilled confirmation form.
  • Power BI MCP joins the catalog; code-review agents can set reasoning effort.
Read full story
Linked Models & Agents:
🤖 GitHub Copilot (Agent Mode / Codex)🌐 github🌐 Microsoft Azure AI
#GitHub Copilot App#/goal#MCP#Power BI MCP#Autopilot#code review
N
Nikunj Handa@nikunjhanda·
👀62
🔥93
View on 𝕏 ↗

OpenAI: GPT-6 Astra Responses API adds async function calling and mid-turn steering

OpenAI engineer Nikunj Handa says GPT-6 Astra ships three Responses API upgrades: async function calling (the model can keep going while tools run), mid-turn steering (inject messages or tool outputs during reasoning), and changing reasoning effort without breaking cache. Details are in OpenAI’s latest-model developer guide.

Key Takeaways

  • Async function calling: model continues while tools run, cutting idle multi-tool loops.
  • Mid-turn steering: inject messages or tool outputs during reasoning without a full restart.
  • Change reasoning effort without breaking cache—useful for cost/latency tuning.
Read full story
Linked Models & Agents:
🌐 OpenAI
#GPT-6 Astra#Responses API#async function calling#mid-turn steering#reasoning effort#OpenAI#API
O
OpenAI@OpenAI·
👀46
🔥97
View on 𝕏 ↗

OpenAI Announces o3-mini Full API Availability: Custom Reasoning Budgets & 200k Context

OpenAI made o3-mini reasoning API generally available with flexible reasoning effort tiers (low/medium/high) at $1.10/1M input.

Key Takeaways

  • Reaches 71.2% on SWE-bench Verified, rivaling human grandmaster algorithm performance
  • Dynamic reasoning_effort parameter balances millisecond completion with deep architectural synthesis
  • Full Prompt Caching support dropping cached token cost to $0.55/1M
Read full story
Linked Models & Agents:
⚡ OpenAI o3-mini (High Reasoning Coding)⚡ GPT-5.6 Sol🤖 Devin 2.0 (Cognition)🤖 Cursor Composer (Agent Mode)🤖 Cline (VS Code Agent)🌐 OpenAI
#OpenAI#O3 mini#SWE-bench 71.2%#Deep Reasoning#Prompt Caching
ADSponsored
D
DeepSeek@deepseek_ai·
👀41
🔥93
View on 𝕏 ↗

DeepSeek-V4-Pro officially launches with variable reasoning effort and native Codex support

DeepSeek officially launched the production-ready V4-Pro model. Featuring enhanced agentic stability in production, it supports low, high, and max reasoning effort toggles, native OpenAI Responses API compatibility, and one-click Codex integration alongside a 50% off-peak pricing tier.

Key Takeaways

  • Optimized for production agent loop stability and precise multi-tool execution.
  • Configurable reasoning effort (low, high, max) balances latency with algorithmic depth.
  • Native OpenAI Responses API support paired with a 50% off-peak pricing schedule.
Read full story
Linked Models & Agents:
⚡ DeepSeek-V4 (Next-Gen MoE)⚡ DeepSeek-R1 (Open Reasoning)🌐 DeepSeek
#Deepseek#Deepseek v4 pro#Codex#API release#Off peak pricing
P
Paul Gauthier@paulgauthier·
👀29
🔥90
View on 𝕏 ↗

Aider Polyglot benchmark: OpenAI o3-pro sets new 85% SOTA for multilingual code editing

Aider creator Paul Gauthier announced that OpenAI o3-pro with high reasoning effort set a new record of 85% on the Aider Polyglot coding benchmark. The benchmark evaluates multi-language real-world codebase editing, solidifying o3-pro's position at the top of practical coding leaderboards.

Key Takeaways

  • o3-pro achieves an 85% solve rate on Aider Polyglot with high reasoning effort, setting a new SOTA.
  • Aider real-repo benchmark remains a gold-standard yardstick for code editing accuracy.
  • Demonstrates an 18%+ leap in unified multi-language refactoring over previous generation models.
Read full story
Linked Models & Agents:
⚡ OpenAI o3-mini (High Reasoning Coding)🤖 Aider🌐 OpenAI
#Aider#O3 pro#Polyglot benchmark#Sota#Code editing#Leaderboard
You've caught up with all stories