Image: OpenAI / techmymoney.com OpenAI Launches GPT-5.6: Three Models, One Framework, New Price Points
OpenAI released GPT-5.6 Sol, Terra, and Luna today — a three-tier model family with a 1.5 million-token context window, an Ultra agentic mode, and prices that undercut Claude Fable 5 at comparable capability.
OpenAI released GPT-5.6 publicly on July 9, 2026. The rollout began with a narrow set of enterprise and government partners, then opened to API and Codex users broadly within hours. ChatGPT access is planned for shortly after.
The release introduces a new naming structure: instead of version suffixes, OpenAI is using Sol, Terra, and Luna as durable capability tiers that can advance independently on their own release cadence.
The three models
Sol is the flagship. It’s built for frontier reasoning, long-horizon agentic tasks, coding, and scientific work. Pricing is $5 per million input tokens, $30 per million output.
Terra targets everyday use. OpenAI positions it as GPT-5.5-competitive performance at roughly half the cost: $2.50/$15 per million tokens.
Luna is the speed and cost play: $1 input, $6 output, still scoring 84.3% on Terminal-Bench 2.1.
The naming frames these as persistent categories rather than point releases. Terra could get model updates without becoming “GPT-5.7 Terra” — similar to how Claude uses code names for major versions.
What’s new
The most significant changes are in context and agentic capability.
Context window: Sol extends to approximately 1.5 million tokens, a 43% increase over GPT-5.5. Training data is updated through May 2026.
Sol Ultra mode: Sol can now deploy subagents to break complex, multi-step tasks into parallel workstreams. This is closer to how Claude Code’s background agents or Anthropic’s orchestration layer works. For long coding jobs, Ultra mode runs sub-tasks concurrently rather than sequentially.
Max reasoning effort: Sol can take extended thinking time on hard problems, similar to extended thinking modes in other frontier models.
Benchmark highlights: Sol hit new state-of-the-art on Terminal-Bench 2.1, the benchmark for command-line workflows requiring planning, iteration, and tool use. On Sol Ultra mode specifically, the score is 91.9%, above Fable 5’s 84.3% and Grok 4.5’s 83.3%. On SWE-Bench Pro, numbers haven’t been officially published as of launch.
Token efficiency has also improved. On benchmark tasks, Sol uses substantially fewer output tokens than GPT-5.5 did on comparable problems.
Price comparison against the competition
This is where the launch gets interesting for teams running large agentic workloads.
Claude Fable 5 is $10 input / $50 output. Sol is half that at $5/$30, with roughly comparable top-end capability on available benchmarks. Terra at $2.50/$15 beats Fable 5’s price significantly while claiming parity with GPT-5.5’s performance.
Grok 4.5 is cheaper still at $2/$6, but Sol outperforms it on most published benchmarks where contamination concerns don’t apply.
For teams that were using GPT-5.5 and upgrading in place, Terra is the clear landing spot: better performance at the same or lower cost. Teams paying Fable 5 prices for frontier work now have a direct alternative to evaluate.
Availability
Initial rollout went to select enterprise and government partners. OpenAI shared the models with U.S. government contacts before the public launch, a pattern the company has followed with previous major releases. Broader API access opened the same day. ChatGPT access is expected soon.
Cerebras deployment is available for select customers during July, supporting up to 750 tokens per second on Sol.
Sources: OpenAI — Previewing GPT-5.6 Sol, OpenAI Developer Community — Introducing GPT-5.6 series, Neowin — OpenAI to release GPT-5.6 Sol, Terra and Luna on July 9, explainx.ai — GPT-5.6: Public Launch July 9