TokenCalculator July 2026 Update: 130+ Models, Hermes-OpenClaw and a Faster Nav

Eight weeks, thirteen frontier-class releases, and a fair amount of catching up on our side. Here is what changed on TokenCalculator this month.

Every New Model, With Real Pricing

We added 23 models covering the June and July release wave: Claude Opus 5, Opus 5 1M and Sonnet 5, GPT-5.6 with the Sol, Mini and Codex tiers, Gemini 3.6 Pro and Flash, Kimi K3.7 with Turbo and Mini variants, Qwen 3.8 and Qwen 3.8 Coder, the full Llama 5 family, DeepSeek V4.5 and R2, Grok 4 and Grok 4 Mini, Mistral Large 3.1 and GLM-5.

We also went back through the existing catalogue and filled in every missing price. Thirteen models previously showed blank input or output rates, mostly older Mistral and Qwen entries. Those now carry real numbers. Several models with an unknown release date have been dated, and two duplicate Llama 4 entries that had been quietly double-counting in the pricing table are gone.

The full catalogue is on the models page, and every entry feeds the token calculator and the price comparison tool.

Hermes-OpenClaw

Our OpenClaw directory has been rebuilt and renamed to Hermes-OpenClaw. The old page tracked one corner of the agentic coding ecosystem. The new one covers the whole thing: terminal agents, AI IDEs, editor plugins, autonomous software engineers, orchestration frameworks and the sandbox providers they run inside.

Forty tools across seven categories, each with a short honest description rather than marketing copy, plus a section on how to actually choose between them. The old URL redirects, so existing links keep working.

Leaderboards and Benchmarks Recalculated

The LLM leaderboard now ranks twenty models rather than sixteen, with TC Scores recalculated across the current field. The benchmarks page was refreshed the same way, and we recalibrated the colour thresholds because scores in the 70s on SWE-bench Verified made the old thresholds meaningless.

Worth reading alongside our piece on benchmark contamination, because the absolute numbers deserve more scepticism than they usually get.

Navigation and Layout

The top bar now uses the full width of the window. The logo sits in the actual top-left corner and the Skills link, Blog link, language switcher, settings and theme toggle sit in the actual top-right corner, instead of both floating inward on wide screens. Two new provider tabs, Moonshot AI and Zhipu AI, were added to the models mega menu, and the footer picked up a Directories column.

Content

Thirteen new articles covering the release wave, from the Llama 5 launch through to the GPT-5.6 counter-punch, plus a decision tree that gives a straight answer for the twelve situations most teams are in. The AI news page was updated through July 21 with the earlier coverage kept in place as an archive rather than overwritten.

The glossary gained new entries for the concepts that came out of this cycle, including context compaction, adaptive effort and benchmark contamination.

Coming Next

Per-model cheatsheets for the new releases, expanded playbooks for agent routing, and a serious pass on page weight. The last one is overdue and we know it.

Full history on the changelog. As always, corrections and requests are welcome through the contact page.