GPT-5.5
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 13
- LiveKit guide: building a production-ready AI voice agent
LiveKit published a guide on taking AI voice agents from demo to production, organized around three questions: whether the agent sounds smart and natural, whether teams can tell how it performs in production, and whether it scales reliably and cost-effectively. Among its recommendations, LiveKit advises measuring the time from pickup to the agent's first word as a separate metric — one unnamed company found about three of every four calls ended at the greeting step — and passing only greeting-level context with the call while loading the rest in the background. It also recommends LiveKit's turn detector (on by default in AgentSession) with adaptive interruption handling, asynchronous tool calls with filler phrases to avoid dead air, noise-cancellation testing, and expressive TTS; for latency it cites LiveKit benchmarks showing Gemma 4 31B at 192 ms time to first token versus 966 ms for GPT-5.5, 1,006 ms for GPT-4.1 and 1,095 ms for Gemini 3.0 Flash, and says some clients moving from GPT-4.1 to Gemma saved 300–500 ms at p50.
LiveKit Blog · 🔥 7 - CAIS dashboard tracks AI job cuts and wage losses by state
The Center for AI Safety (CAIS) maintains a Wage Loss Dashboard that maps state-level AI-attributed layoffs and converts them into estimated annual wage losses. It counts just over 117,000 AI-attributed job cuts in the US so far this year — about $8.5 billion in estimated annual wage losses, or roughly $25 per person nationally (about $56 in California and $14 in Alaska) — drawing on Challenger, Gray & Christmas layoff announcements, Bureau of Labor Statistics wage data and Census population estimates. The project also cites the Remote Labor Index it co-developed, saying the best AI agent rose from completing 2.5% of real-world freelance projects at professional quality to 15.8% eight months later, while the dashboard excludes jobs created by AI, productivity gains and re-employment.
ZDNET AI · 🔥 6 - Microsoft study: agent long-term memory may hurt; Fork-and-Flush resets context
A Microsoft Research paper posted to arXiv on October 5, “Fork-and-Flush: Escaping Idea Basins in Autoresearch Agents,” reports that autoresearch agents can get stuck in “idea basins”: running the same agent on the same task eight times yielded final scores from 68.3 to 100.0 (median 81.9), and throwing extra compute at the weak runs did not close the gap. The proposed Fork-and-Flush method copies the agent into parallel branches that inherit the full workspace — code, files and experiment results — but start with a cleared chat history, then keeps the best-scoring branch after a fixed period. Across 13 long-horizon tasks under the same compute budget, the normalized average score rose to 0.78 versus 0.47 for a single run and 0.54 for best-of-four, with standard deviation falling from 0.26 to 0.16; in ablations, flushing alone gave roughly a 22% gain while forking and flushing together worked best. Tests with Qwen3.5 at 4B, 9B and 27B on MAX-3-SAT showed run-to-run variance peaks near the model’s capability edge, at about 0.5 attainment, suggesting the effect may persist as models improve.
36氪 人工智能 · 🔥 6 - OpenAI retires GPT-5.5 from ChatGPT and Codex as older APIs face sunset
On October 14, OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work and Codex sessions signed in with a ChatGPT account, while API access and API-key Codex sessions remain unaffected. Specified versions of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini are also counting down to discontinuation on January 14, 2027, leaving developers roughly three months to migrate; OpenAI advises checking model settings on older tasks and switching to replacement versions. Reports say GPT-5.5 shipped in April 2026 and was superseded by GPT-5.6 and GPT-6 in under six months, and some Reddit users have posted prompt-writing guides to carry its tone and habits into private API setups, citing MentalHealthBench scores — user claims rather than official statements.
36氪 人工智能 · 🔥 7 - MIT study: AI theory problem-solving jumps from 2% to 35% in two years
Researchers at MIT — Neil Thompson, Gabriel Manso and Claudio Mayrink Verdun — tested OpenAI's o1, o3, GPT-5, GPT-5.1, GPT-5.4 and GPT-5.5 on theoretical problems in probability, statistics theory, economic theory, high-energy theory, gravitation and cosmology, and quantum physics. According to the study, the best models' share of correct solutions rose from about 2% to about 35% between mid-2024 and mid-2026. The problems came from papers posted between December 2025 and May 2026 whose solutions appeared after the training cutoff of GPT-5.5 and earlier models, and internet access was blocked during testing, so the models had to derive answers themselves. Thompson says many theory questions could be "totally transformed" within a few years, while the authors caution that humans are still needed to generate ideas and verify solutions; OpenAI withdrew several solutions from its latest dataset after publication.
WIRED AI · 🔥 0 - Berkeley and Sakana AI propose MASS multi-agent self-supervision
Researchers at UC Berkeley and Sakana AI propose Multi-Agent Self-Supervision (MASS), in which the same model alternates between task-solving agent, workflow optimizer and evaluator: it first optimizes how multiple copies of itself collaborate with weights frozen, then distills the resulting trajectories back into shared weights. Running two MASS rounds on Qwen3.6-27B inside the qwen-code environment raised its win rate against the base model on held-out test tasks from 53.9% to 69.9%, and lifted per-output-token scores on MLR-Bench, DSBench, AstaBench E2E-Hard and ScienceAgentBench to roughly 1.2–1.6x the base model, while Terminal-Bench 2.0 and SWE-bench Verified came in at about 0.96x and 0.94x respectively. The paper also reports the evaluator's agreement with external judges rising from 0.73 to 0.93 and better training efficiency from multi-agent trajectories than single-agent ones; GPT-5.5 and Claude Opus 4.8 served as external judges in the experiments.
36氪 人工智能 · 🔥 0 - REA tops GitHub as AI takes over reverse engineering
The open-source project REA (Reverse Engineer Anything) plugs disassembly tools such as Ghidra, IDA and Hopper into AI coding agents like Claude Code, Codex and Cursor through MCP, so a user can ask an agent to investigate how an app's feature works and show the evidence. Star History counted 78,800 new stars for REA in the week of Oct 4–10, first among GitHub projects that week, with its total nearing 100,000 when the piece was written; a homepage case traces a DX-Ball sound call to the function that computes left and right channels, reconstructs it in C, passes 3,205 tests and compiles to 63 bytes identical to the original. Elias Bachaalany, co-author of a reverse-engineering textbook, dismissed it with the single word 'hype', and in the same week developer Brandon Thomas released ArtCraft, seven free Rust programs built with Claude Opus 5.5 that target Adobe Photoshop, Premiere and others; PhotoCraft gained 42,900 stars in a week, but users reported broken PSD files containing Smart Objects, crashes when importing many clips, and only 5 of 30 filter-effect tests matching.
36氪 人工智能 · 🔥 0 - AesCode: Aesthetic Code Generation with Decoupled Cross-Modal Rewards
An arXiv paper introduces AesCode, a method that generates code grounded in a content prompt while using an image generated from that same prompt as aesthetic guidance. It represents the target as a design graph covering the whole canvas so each property is attributable, then derives complementary cross-modal rewards: deterministic verifiers score properties parsable from code and rendering, while a VLM judge assesses non-parsable ones using a sample-specific Visual Graph Rubric tied to graph elements and relations, optimized via reinforcement learning with per-channel normalization. The authors also build a benchmark scoring verifiable correctness and visual quality separately, and report that AesCode-8B and AesCode-32B improve visual quality by 31.1 and 22.4 percentage points over same-scale Qwen3-VL baselines given the same references, reaching scores comparable to closed-source frontier models such as GPT-5.5 and Claude Opus 4.8 (author claims).
Hugging Face · Papers · 🔥 0 - Sakana AI's MASS lets a model self-improve, but blind spots compound
Researchers from Sakana AI and UC Berkeley introduced "Multi-Agent Self-Supervision" (MASS), in which a single language model acts as solver, evaluator and optimizer to improve itself recursively without a stronger teacher model. Using Qwen3.6-27B with Qwen Code, its win rate against the base model rose from 53.9 percent after one round to 69.9 percent after two, the self-evaluator's agreement with external judges climbed from 73 to 93 percent, and it reached 1.2 to 1.6 times the base model's score per output token on MLR-Bench, DSBench, ScienceAgentBench and AstaBench. The paper warns, however, that the internal evaluator sometimes accepted manipulated or invalid results as successes and baked them into the next model, a shared blind spot the authors call the potential "Achilles' heel" of recursive self-improvement and a reason to demand independent checks, limits on autonomy and clear conditions for intervention.
The Decoder · 🔥 0 - Microsoft study: clearing chat history helps AI research agents find better solutions
A Microsoft Research study finds that long-running AI research agents plateau because accumulated conversation context traps them in narrow "idea basins" within the solution space. In eight independent runs of the same agent on the same task, final scores ranged from 68.3 to 100.0, and the gaps persisted even with substantially more compute. The proposed Fork-and-Flush method splits the agent into parallel branches that inherit the workspace and files but lose all chat context; across 13 long-running tasks (including polyomino packing, job-shop scheduling and cryptanalysis) it reached a min-max-normalized average score of 0.78 versus 0.47 for single runs and 0.54 for best-of-4 — relative gains of 66.0% and 44.4% — with normalized standard deviation falling from 0.26 to 0.16. Flushing context without branching alone still improved results by about 22%; the agents used GPT-5.6-terra and GPT-5.5 through the GitHub Copilot CLI, with individual runs lasting hours to over 100 hours.
The Decoder · 🔥 0 - RoboFoundry freezes base models and evolves the embodied agent system around them
A team from Nanyang Technological University, Beihang University, National University of Singapore, Shanghai Jiao Tong University and Yundie Technology proposes RoboFoundry, an embodied agent framework that formalizes the agent system itself as a "Self-Evolving System-as-Policy": base model parameters stay frozen while evolution happens in the context system, skill system and execution interface around the model, driven by an execute-reflect-repair-promote loop and a shared semantic interface for reuse across robots. On EmbodiedBench all five base models improved, with GPT-5.5's average success rate rising from 56.9% to 72.7%, above unevolved GPT-6 Astra at 71.9%; on the long-horizon memory benchmark RoboMemArena it reports task success at least 39.0% above all baselines, and it tops LIBERO-PRO perturbation tests at 91.9% average success. Real-robot experiments on an AgileX platform reached 70% success on a nesting-cups task versus 30% for π0.5 alone, alongside zero-shot skill transfer to unseen towels, semantic navigation on a Unitree G1, and online evolution after deployment.
机器之心 · 🔥 0 - LegalOn halves Codex costs by tiering GPT-6 Astra, Sol and Luna
After embedding Codex in its development process, LegalOn Technologies initially gave engineers unlimited access to GPT-5.5 in Fast mode, which pushed up spending. Its AI-powered Development CoE then introduced model-selection guidelines: GPT-6 Luna for everyday implementation, GPT-6.1 Sol for standard design, data analysis and documents, and GPT-6 Astra for complex analysis, architecture and agent orchestration, alongside restricting Fast mode by default and setting monthly budget caps per department, group and individual. OpenAI says the company cut estimated daily costs by roughly 65% versus GPT-5.5 while maintaining development speed, with mature businesses asked for about 20% better cost efficiency and new businesses given generous budgets.
OpenAI · 🔥 0 - MIMESIS: a 9B user simulator as a training environment for agents
The paper introduces MIMESIS, a purpose-built 9B user simulator trained on human conversations with explicit reasoning supervision and 13 behavioral patterns derived from real user interactions. The authors report a SOUL-Index of 65.7, above the strongest frontier model, and say it improves behavioral fidelity by 13.4 points and reduces Turing distance by 3.6 points versus the strongest baselines, Claude-Opus-5 on RealUserSim and SimulatorArena respectively. They then freeze the simulator to train an agent with multi-turn reinforcement learning, reporting better results than training with GPT-5.5 across nine unseen user simulators, and propose Coached On-Policy Self-Distillation (CSD) to turn the simulator's private reasoning traces into token-level supervision. Code is linked in the paper.
Hugging Face · Papers · 🔥 0
Experience and discussion from the community
Share my GPT-5.5 experienceAsk about GPT-5.5
Nobody has shared their experience with GPT-5.5 yet.